The Impact of External Support on Intrastate Conflict

Introduction

One of the difficulties of waging a successful rebellion against a state is the asymmetry of the resources both sides can effectively mobilize. Often rebel groups will turn to asymmetrical strategies such as guerrilla warfare to compensate for disparities in resources. However, rebels will either need to defeat the state outright or make the conflict exceedingly costly for the state in order to achieve victory. Support from third parties can help rebel groups reduce the asymmetry between themselves and the government. Rebels may not receive the high-end resources a state may possess, but food, medicine, small arms, and potentially direct military support can supplement domestic resource mobilization and contribute to a more successful rebel movement.

The importance of this puzzle is the extent to which external support is an effective method to shape the outcome of an intrastate conflict. States often provide support to foreign rebels to cheaply achieve policy goals such as removing a hostile regime through domestic actors. The degree to which providers and recipients of external support evaluate the efficacy of external support for the outcome of intrastate conflicts shapes policy decisions and can thus have broad policy implications for how they achieve their policy goals. For example, a state may pursue a détente-like strategy with an opposing state rather than try to undermine them by supporting rebels if external support is considered ineffective. With a lower likelihood of success, states will consider support to a side of an intrastate conflict less as one of their primary tools to achieve goals and will focus on alternative strategies.

The goal of this paper is to assess the extent to which types of external support impact intrastate conflict. It is expected that external support to rebels should have a small but meaningful impact on the likelihood of a rebel faction to win. This should be particularly more pronounced at the higher levels of support (ex: direct troop deployment) as the higher levels of support should provide rebels with more capabilities to fight the state. This relationship should be identical for external support to states involved in an intrastate conflict, but with a smaller magnitude. The smaller magnitude is likely because external support supplements a state's capabilities like external support to rebels, but that support will likely be marginally less effective than the equivalent support to rebels. While the various domestic factors should take priority, external support can be a supplement for the capabilities of both sides and should thus have some impact on the outcome of intrastate conflicts.

Review of Statistical Analyses of External Intervention

The current literature of statistical analyses on the impact of external intervention in intrastate conflict is insufficient. The common way that intervention is operationalized is by the direct military involvement of a third party state. Among common datasets, this is the criteria for external involvement or internationalization of a conflict (Sarkees & Wayman 2010) (Gleditsch et al. 2002). Nonetheless, there are quality efforts to assess the role of external intervention through more nuanced statistical models of external support to participants of civil wars. Although this is not where the literature should be, it is moving towards a more satisfactory analysis. Even so, earlier attempts focusing on direct military intervention can portray some, but not all, dynamics that this paper examines because the literature generally examines external support at a much higher threshold of involvement.

The definition of a proxy is somewhat contentious due to the inherent ambiguity of proxy support. While there are obvious cases such as Soviet and Chinese support to North Vietnam, there are also smaller-scale proxy relationships such as those with Cuban and Latin American rebel groups. The traditional definition established from Cold War proxy conflicts is that there is an asymmetry in power and that the lesser power relies on the larger power for the capability to act (Duner, 1981, 356). Generally this relationship would create the expectation that the weaker power is submissive to the interests of the larger benefactor. This definition likely exaggerates the influence of the larger party on the lesser power. Withholding resources can be an effective way to induce compliance, but reducing the effectiveness of the lesser power weakens the benefactor's ability to undermine the opposing actor. There is some overlap with an alliance and a proxy relationship in that both are coordinating towards a desired goal or outcome, and that some aid can flow between allies. However, allies do not become reliant on the aid, which makes the relationship less submissive. A proxy likely has a freedom of action much like local actors in the cleavages model described by Stathis Kalyvas (2013, pp. 475) because the benefactor cannot micromanage the proxy while maintaining minimum expenditures.

Dylan Balch-Lindsay and Andrew Enterline (2000, pp. 630) evaluate the impact external intervention has on the duration of a conflict. They find that interventions generally increase the duration of the conflict with a "balanced intervention" having the largest impact on duration. Their analysis is important for two reasons. Firstly, their control variables, particularly "government strength," suggest that domestic factors are important to control for in order for a defendable statistical analysis of external support (Ibid). While domestic variables are important, intervention is still significant. Nonetheless, regardless of domestic factors, it is reasonable to expect that some conflicts, especially the Vietnam War and Soviet-Afghan War, would have ended years earlier if not for external investments in those countries. Secondly, to reiterate, the use of a high threshold of external support excludes any commentary on lower levels of support that are more frequent.

Contrary to Balch-Lindsay and Enterline, Paul Collier, Anke Hoeffler and Måns Söderbom (2004, 261) argue that intervention can reduce the duration of a conflict. Their research on the duration of civil wars found that support only to rebels statistically lowered the duration of the conflict. This is expected to some degree because supplies to a rebel group are marginally more valuable and useful to the rebel group than additional support is to a government that should already have better resource mobilization. Furthermore, like BalchLindsay and Enterline, they assess the impact of military interventions, but also account for economic interventions (Ibid). While the expansion of types of external involvement is a step forward, it does not elaborate on the potentially effective low-cost support that takes the form of material aid short of direct military intervention.

Lotta Jarbom and Peter Wallensteen's work with the UCDP dataset highlights the impact of external support on the outcome of a civil war. They suggest that continued external support to only one side tends to result in a ceasefire or victory and that those settlements involving external support are more durable than without external support (Harbom & Wallensteen, 2005, 629). Their work also emphasizes the importance of neighbors as intervening third parties which helps narrow the list of potential intervening states to reduce type two errors. Their inclusion of "secondary non-warring support" is an improvement in expanding the variation in support, but it can be further expanded (Ibid). Their work suggests that there is sufficient variation for outcomes in the UCDP dataset.

Patrick Regan and Aysegul Aydin (2006, 748) try to elaborate on the variation in outcomes and duration of the conflict by assessing "diplomatic intervention" as a way in which third parties can end the conflict. According to them, diplomatic intervention, such as economic intervention, improves the chances of the conflict resulting in a peace settlement, but the intervention declines in effectiveness over time. While their research emphasizes a different type of intervention than what is addressed within this essay, it is a useful way to explain how intrastate conflicts end; at some point there must be diplomacy and a third party can help speed that process up, whether by providing a neutral space to meet or incentives for each side to meet at the negotiating table. Their analysis goes beyond the scope of this essay, but it does reflect dynamics beyond relative capabilities and resources that are often missed in quantitative studies.

Tyrone Groh (2010, 7) evaluates the degree to which proxy war can be an effective low-cost option in his dissertation. While he is right to point out that the degree to which the external benefactor controls the beneficiary is important for cost-benefit calculations, this aspect is beyond the scope of this essay. More importantly, he evaluates the cost of proxy support primarily in terms of material costs, influence, and reputation costs at the international and domestic level. While he categorizes costs and benefits, none of these variables represent the outcome of the conflict. Nor does it account for benefits and costs to interstate rivalries that appear in the literature as a reason to provide support (Ibid, 254). Nonetheless, his dissertation provides other benefits and costs outside of this paper which will help provide context for the results and implications.

The general limit of this body of research is that the lower levels of intervention that are far more common are not analyzed. The threshold of intervention for statistical analysis is high enough to omit a significant number of cases of external support that can improve the evidence for assessing the utility of intervention. In addition to being able to cover cases that would not have been considered external support under the usual dichotomy of military intervention, the treatment of external support as an ordinal variable allows for a greater variation in the independent variable in order to more accurately depict the influence of external support on a conflict. To address this limitation and contribute to the discourse, this paper treats external support as a more expressive variable that accounts for the variety of ways in which external support can exist.

Hypotheses

The primary hypothesis that this paper tests is that higher levels of external support to rebels increases the probability that the rebels will win and that the state will lose. External support to rebels as a whole should impact the outcome because it should allow them to improve their capabilities and more effectively resist and fight the government. The higher levels of external support, such as deploying troops, should have a greater impact because they represent a greater investment than lower levels. External support can provide the resources to help rebels mobilize manpower. Intangible incentives to mobilize individuals, such as promises, are not accounted for in this analysis (to the extent that it is not captured indirectly by control variables). At the very least, a soldier may fight for a cause, but at the end of the day he needs food, water, medicine, and a gun to effectively fight the state. Another state can provide that type of support.

The secondary hypothesis that this paper tests is that different levels of external support to states increases the probability that states win and that the rebels lose. This is expected because the state's capabilities should improve from additional supplies and/or reinforcements. It follows the same logic as rebel support, but it is expected that the state should benefit less than rebels do from support because external support to a state is likely a smaller portion of a state's total capabilities than the portion of external support to rebels is. A state should have the means to create and maintain a force to combat rebels in most cases. While not accounted for, the magnitude of support (within each category) may be larger than the support to the rebels. This is likely the case because, while rebels would not likely benefit from complex weaponry that is difficult to maintain and support, a state might be provided with complex weaponry since it likely has access to supporting infrastructure. Despite states likely receiving greater support than rebels, the support is probably a smaller percentage of total strength and should therefore have less of an impact.

Variables

The independent variable is the level of external support for rebels. The variable consists of five values: 0 is no support, 1 is alleged support, 2 is nonmilitary support (medicine, food, etc.), 3 is military support (guns, munitions), and 4 is direct military support (troops). This scale is also used for the level of external support to the state. Alleged support is the lowest level of external support because it is assumed that the support, regardless of type, is small enough to evade verification and thus have a lower impact on the conflict. It is assumed that each higher category costs more to supply because each section is more expensive financially and politically in general. However, it is entirely possible for nonmilitary support to be more expensive than military support or that a small troop deployment is cheaper than arming a rebel group. These cases should be rarities and should not impact the results to a significant degree. It is also assumed that the higher levels include the lower types of support because if a state is willing to support at a higher level, it should also be willing to provide cheaper forms of support. If noncombat supplies are needed, it would make sense for them to be sent in addition to military supplies.

Alleged support is included as a variable because cases where support is suspected should not be included in the no support category, as these cases would contaminate the no support variable. Likewise, including it within the other levels of support would potentially contaminate those as well. While the number of cases is small (alleged support to the state was removed because there were so few cases), coding them as the other values would not be appropriate. There is also the potential that alleged support can impact the outcome through perceptions about rebel capacity. Alleged rebel support without actual support would more likely encourage a stronger state response that would defeat the rebels.

The dependent variable is the outcome of the conflict. The variable consists of three values. 0 is the absence of victory, 1 is rebel victory, and 2 is state victory. All outcomes were considered a neutral outcome if the outcome did not have a victorious side or did not terminate yet. It may not be appropriate to include vague types of termination such as "other" and "no or low activity," but these were included to ensure a larger sample size and can be considered an absence of victory so it is appropriate. However, anti-colonial cases were removed because their outcomes were coded as "end of anti-colonial conflict," which does not adequately describe the outcome for use in this analysis. While less descriptive than the Correlates of War outcome variable, the outcome variable used in this analysis allows for a larger sample size in each category to help find potential statistical significance.

The control variables consist of other ordinal or categorical variables such as conflict type, relative strength between rebels and government, the rebel's internal capacity to recruit and acquire arms, and the existence of other non-state sources of external support. The control variables were chosen as other factors that should impact the outcome, notably the type of conflict and the rebel's domestic capability relative to the state.

Methodology

The data used for this paper are taken from the Non-State Actor Data dataset compiled by Cunningham, Gleditsch, and Salehyan as an expansion of the UCDP dataset on civil conflict (2012). While retaining the structure of the UCDP dataset, the expansion includes variables reflecting different relative capabilities in addition to a more expressive external support variable than the UCDP dataset on civil conflict. Due to problems found with coding in other datasets, the external support variables were checked and slightly altered depending upon whether information contrary to the coding was found. Few cases were altered. This dataset provides a better independent variable with numerous control variables to account for general differences in domestic dynamics.

The data is analyzed through a series of logit models for rebel and state victory. This is appropriate because the goal of this paper is to discuss the average impact of external support to rebels and states on the outcome of the conflict. While a linear probability model may be similar, it does not adequately represent the limits of the impact of independent variables (under 0 and over 1). A logit or probit model is necessary to properly represent the probability distribution. While either model would work, the logit model is used out of convenience. The analysis is accomplished by running two nearly identical series of logit models side by side where each victory category is compared to the absence of victory. Each series consists of five models with each model including additional control variables. The marginal effects are reported at the mean because that is the best way to acquire meaningful values to analyze.

The independent variables were aggregated into binary variables because an ordinal scale would not be appropriate for analysis. The difference between nonmilitary support and military support is not the same as the difference between no support and alleged support. This logic also applies to control variables that are ordinal to more effectively measure their impact as a whole. While less concise, this method allows for specific relationships to be presented and avoids misleading trends.

The control variables were added in sets by the type of variable. For model two, external support to the state is added because rebel support and state support are likely correlated and the impact of government support would be captured by the rebel support variables without their inclusion. Model three added the types of conflict (baseline being civil war). The ways in which different conflicts are fought is different and the goals often vary. A terrorist campaign would emphasize suicide bombings and indiscriminate strikes against citizens, while a civil war is often open combat with an emphasis on securing the capital. These factors affect the inherent probability of success and are necessary to control for.

Model four added relative domestic capabilities of the rebel group (baseline being parity). While relative capabilities likely captures some of the impact of external support, without a more detailed description in the codebook, the degree to which the two sets of variables overlap in coding is uncertain. It is still better to use the relative capability variables to account for domestic differences because while there is likely overlap in explained variance, relative capabilities capture other domestic factors that are necessary to control for. With external support considered generally a supplement, domestic mobilization should represent a significant portion of state and rebel capabilities.

Model five added territorial variables (baseline is none) and non-state actor support (baseline being no support). Alternate sources of support (militias and smuggling networks) can potentially impact the conflict to the degree that state support could, depending on the magnitude. A rebel faction's presence domestically and/or abroad can potentially be an indicator of their capacity to fight the state. If they control territory, they can more easily mobilize the resources of the territory under their control. Safe havens provide a sanctuary to prevent the collapse of a rebel group and can thus impact the outcome as well. States may only provide money as external support which would require access to a smuggling network to provide the resources necessary to fight.

Results

The results suggest that external support for rebels and states has a minimal impact on the outcome of intrastate conflict. Hypothesis one, that higher levels of rebel support would increase the likelihood of a rebel victory, is unsupported. Models three, four, and five either present the various types of rebel support as statistically insignificant or with a small magnitude of 1-2%. Furthermore, some of the rebel support categories that are statistically significant, such as alleged and nonmilitary rebel support in model four, are negative, indicating that additional assistance reduces the likelihood of a rebel victory.

Hypothesis two, that higher levels of state support would increase the likelihood of a state victory, is also unsupported. Similar to hypothesis one, the majority of state support variables were either statistically insignificant or have a negligible magnitude. The one significant type of state support is nonmilitary state support which has a 19% probability of a state victory in model four. While this piece of evidence provides partial support to hypothesis 2 in that nonmilitary aid improves the likelihood of state victory more than no aid, the lack of significance for military aid and troop support also means that a conflict with a lower level of external support to the state relative to troop or military support has a greater likelihood of state victory.

The significance of alleged rebel support in analyzing the likelihood of a state victory in models three and four is another unique result among the external support types. Alleged external support is the most ambiguous category in an analysis of inherently secretive behavior and is the least homogenous category. Alleged support could range from there being no support to covert troop deployments. Therefore this unique result should not be considered evidence that external support has an impact on the outcome of the conflict without concurrent results for the other categories of external support. With the absence of additional evidence, it is far more likely that conflicts with alleged rebel support have some unique characteristic related to the greater ambiguity in those conflicts that reduces the likelihood of a state victory.

While the reduced sample size for models four and five can be a reason that the external support variables fail to remain significant, these models capture alternative explanations of variance in the outcome. The relative capability variables added by model four are particularly useful. The additional explained variance of the last two models is demonstrated by the increase in the pseudo R^2 values for each model which justifies the additional control variables. The statistically significant results across models three, four, and five indicate that some types of conflict (coups) have a greater capacity for one side to win while other types of conflict (ethnic and Islamist) more often lead to an absence of victory.

A noteworthy result of model five is the statistically significant impact of rebels having an external presence on state victory. The results indicate that while a rebel group with an external presence can reduce the probability of a state victory, it does not significantly improve their ability to win the conflict. Although the magnitude is about -2.5% for extensive external rebel presence and some rebel external presence, the percentage can still be considered a relevant result considering the plethora of factors involved in an intrastate conflict. This result can be logically explained through the increased difficulty a state has in removing a rebel group if the rebels can retreat to a safe haven across the border.

Another interesting dynamic is that a state is 6% less likely to win if it receives minor non-state actor support. This is likely the case because states that need to rely on non-state actors to fight rebels likely have a poor capacity to mobilize resources and poor governance capabilities. This variable can be an indicator that the state is desperate enough to rely on militias in order to combat the rebels.

While extraterritorial might appear to be a significant variable with a 98.5% increase in the probability of a state victory, this indicates the small number of extraterritorial conflicts rather than serving as a strong indicator of state success. The logic that would be needed to justify the results is not as compelling as the logic for the difficulties of a state victory in extraterritorial intrastate conflict. Fighting further away from a power base should make it more difficult for a state to win due to the difficulties of logistics, power projection, and mobilizing local support.

Conclusion

This analysis suggests that higher levels of external support in intrastate conflict to rebels and to states do not significantly impact the outcome of intrastate conflicts. What appear to be more important for the outcome of an intrastate conflict are domestic variables, especially the type of conflict. Domestic variables may account for the inherent selection bias of which states and rebels receive external support because states and rebels with an inherent high probability of success may receive support because they are already likely to succeed. While the statistical insignificance of external support is contrary to previous research on the impact of external support on intrastate conflict outcomes, the results encourage skepticism regarding the role of external support rather than an outright contradiction of other analyses.

This analysis should be considered an initial attempt to discuss the importance of external support for intrastate conflict using a more expressive external support variable. Within the dataset, more concise models could yield more significant results. This would involve reducing the number of control variables while condensing ordinal scales. Filling in the gaps in the dataset could minimize the decline in the sample size across models. However, the key to improving this analysis is to increase the sample size by including more cases. The larger sample size should allow for the more inclusive model to find more statistically significant variables because the standard error would be smaller. While it would be difficult to research all the variables used in this analysis for pre-World War II cases, this dataset can be used to further experiment with models and make the necessary variables more explicit to simplify the expansion of the dataset to earlier cases.

In regards to the policy implications of this analysis, states should be more skeptical about their capacity to impact intrastate conflict through different types of external support. While the results indicate a limited impact on the outcome, there are other reasons to provide external support that may make it worthwhile that are not accounted for in this analysis. As some of the literature has suggested, external support can lengthen the conflict, which may be in the interests of the support providers (BalchLindsay and Enterline, 2004, pp. 475). Support could be a symbolic gesture to improve relations between the provider and recipient. In addition, there could be issues of credibility and future expectations, especially during the Cold War, that incentivize states to support rebels or states in retaliation to events such as regime change and interference by a rival state. In conclusion, states should continue to provide external support to participants in an intrastate conflict so long as it is considered beneficial, but its cost-benefit calculation should contain a low probability of successfully impacting the outcome of the conflict.

Author

Adam Parente was a student of International Studies and Economics. He graduated in May 2014. School of International Service (SIS), American University.

References

Balch-Lindsay, Dylan, and Andrew Enterline. "Killing Time: The World Politics of Civil War Duration, 1820-1992." International Studies Quarterly 44 (2000): 615-42.

Collier, Paul, Anke Hoeffler, and Mans Soderbom. "On the Duration of Civil War." Journal of Peace Research 41, no. 3 (2004): 253-73.

Cunningham, David, Kristian Gleditsch, and Idean Salehyan. "Codebook for the Non-State Actor Data." University of Essex Center for the Study of Civil War (24 January 2012): 1-8.

Duner, Bertil. "Proxy Intervention in Civil Wars." Journal of Peace Research 18, no. 4 (1981): 353-61.

Gleditsch, Nils Petter, Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg, and Håvard Strand. 2002. "Armed Conflict 1946-2001: A New Dataset." Journal of Peace Research 39(5).

Groh, Tyrone. "War On the Cheap? Assessing the Costs and Benefits of Proxy War." PhD diss., Georgetown University, 2010.

Harbom, Lotta, and Peter Wallensteen. "Armed Conflict and Its International Dimensions 1946 2004." Journal of Peace Research 42 (2005): 623-35.

Kalyvas, Stathis "The Ontology of ‘Political Violence': Action and Identity in Civil Wars." Perspectives on Politics 1, no. 3 (22/06/2013): 475-94.

Regan, Patrick, and Aysegul Aydin. "Diplomacy and Other Forms of Intervention in Civil Wars." Journal of Conflict Resolution 50, no. 5 (2006): 736-56.

Salehyan, Idean, Kristian Gleditsch, and David Cunningham. "Explaining External Support for Insurgent Groups." International Organization 65, no. 4 (07 October 2011): 709-44.

Sarkees, Meredith Reid and Frank Wayman (2010). Resort to War: 1816 2007. CQ Press.