We Didn’t Start the Fire: The Effect of Foreign Sponsorship on Peace Duration

Abstract: 

Previous literature addressing the phenomenon of civil war recurrence has examined the domestic sources of recurring civil wars, but have not adequately assessed the impact of foreign sponsorship on civil war recurrence.  The literature exploring foreign sponsorship of civil wars has examined the effect of foreign support on conflict termination, conflict intensity, and conflict outcome, but has not sufficiently explored the impact of foreign sponsorship on conflict recurrence.  This project seeks to fill these gaps and argues that foreign support plays a key role in determining peace duration by increasing the capacity of potential combatants to engage in war. This project further contributes to the literature on foreign support by examining the influence of offensive, defensive, and ambiguous types of support.  Using Cox proportional hazards models, this project finds that offensive types of support decrease the duration of peace (increase the probability of civil war recurrence), while defensive types of support increase the duration of peace (decrease the probability of civil war recurrence). This study provides strong evidence for the importance of distinguishing different categories of foreign support—distinct categories of support have fundamentally different effects on peace duration.

Table of Contents: 

    Introduction

    Between 1996 and 1997, Zaire experienced a bloody civil war pitting the Mobutu government against rebel forces—led by Laurent-Désiré Kabila and supported by the governments of Rwanda, Uganda, and Angola.  The foreign-sponsored rebellion was successful, and the newly empowered Kabila renamed the country the Democratic Republic of the Congo (DRC).  Kabila wanted to ensure his new country would be more than a mere pawn of the foreign states that helped place him in power, and dismissed foreign military forces (most of which were Rwandan) from the DRC.  By 1998, a second civil war had broken out, as rebel forces—again backed by Rwanda and Uganda—began contesting the authority of the fledgling DRC government.  Ultimately, the Kabila government emerged victorious after receiving support from Namibia, Zimbabwe, Angola, and Chad against the rebel forces supported by Rwanda and Uganda.  The “Second Congo War” took place barely a year after the “First Congo War.”  The Second Congo War is estimated to have resulted in the deaths of 5.4 million people (and the displacement of millions more), making it the bloodiest conflict since World War II.  The name of the country may have changed, but conditions of brutal foreign-sponsored civil war did not.

    Recurring civil wars—such as the Congo Wars—are a chronic source of violence, suffering, and pain for many people.  For some regions of the world, the cycle of violence seems inescapable—a condition known as the “conflict trap” (Collier and Sambanis 2002; Collier et al. 2003).  In many cases, the violence within the borders of a state are heavily funded and supported by actors outside the state—such as Rwanda and Uganda in the Congo Wars.  This phenomenon is the impetus for this project, which seeks an answer to the following questions: What role does foreign sponsorship play in civil war recurrence?  Why are some cases of foreign support capable of interrupting peace, while others are not?

    This project seeks to fill two major gaps in the literature regarding civil wars.  First, while studies have examined the role of foreign sponsorship in civil war, empirical studies have not operationalized external sponsorship as a primary explanatory variable. Second, most work focusing on foreign sponsorship has investigated its effect on civil war onset or civil war termination, while neglecting a full exploration of sponsorship’s impact on civil war recurrence.  Furthermore, this project has relevance for contemporary global affairs. The conflict trap is a significant source of human suffering, and the contribution made by this project will allow better prediction of civil war recurrence. This improved predictive capability will better empower policy-makers and peacekeepers to prevent such violence before it occurs. 

    The remainder of this project will be organized into the following sections: 1) literature review; 2) theory and hypotheses; 3) research design; 4) results and analysis; and 5) policy implications and avenues for future research.

    Literature Review

    This literature review will be organized into two major sections: 1) a review of literature addressing foreign sponsorship in civil wars (my independent variable of interest); and 2) a review of literature addressing civil war recurrence (my dependent variable).

    Foreign Sponsors

    A “foreign sponsor” is an actor located external to a country experiencing civil war that is providing some form of support to a warring party within said country. There exists a wide variety of literature concerning the role of foreign sponsors in civil war. This literature explores such topics as what types of rebel groups receive foreign sponsorship, how foreign sponsorship can increase the likelihood of interstate conflict, and the roles of ethnic ties in sponsorship (Gleditsch 2007; Salehyan, Gleditsch, and Cunningham 2011; Salehyan 2008).  In a seminal article in the scientific study of civil wars, Fearon and Laitin found that foreign sponsorship significantly increased the probability of civil war onset (2003). This is intuitive, given that rebel groups are typically funded by donations from foreign states and diasporas, not by the population of the warring country (Collier et al. 2003). Despite widespread interest in the role of foreign sponsors, inadequate attention has been paid to the role foreign sponsors play in civil war recurrence and the different types of support sponsors can provide.

    The various types of support sponsors can provide has received little attention in the quantitative literature. There has been some notable work, however, that disaggregates general support into its primary subcategories in the qualitative literature. Perhaps the most systematic attempt was undertaken by Byman et al. (2001), who not only divided foreign support into ten categories, but also established a hierarchy for understanding the varying degrees of significance these categories of support have in bolstering an insurgency. This study places types of support into three tiers: “critical forms of support” (safe haven and transit; financial resources; political support and propaganda; direct military support); “valuable forms of support” (training; weapons and materiel); and “minor forms of support” (fighters; intelligence; organizational aid; inspiration) (Byman et al. 2001). While such a detailed disaggregation is important to the establishment of a thorough understanding of the process that connects foreign sponsorship and civil war recurrence, Byman et al. (2001) derive their hierarchy from a series of case studies, and do not quantitatively test its validity.  Other work, however, has tested the significance of safe havens, financial support and training (Fearon and Laitin 2003). Adamson (2013) provides a more parsimonious division of the mechanisms by which a sponsor can influence a civil war.  While her mechanisms were established with the specific purpose of analyzing the way diasporas can influence civil war, there is no reason this classification scheme could not be applied to other categories of sponsors. Foreign states, for example, can also engage in resource mobilization and lobbying-persuasion. The two mechanisms present in Adamson’s analysis are resource mobilization and lobbying-persuasion.  Resource mobilization would include numerous categories present in the Byman et al. (2001) typology: financial resources, direct military support, weapons and materiel, fighters, intelligence and organizational aid.  Lobbying-persuasion essentially captures the Byman et al. (2001) category of political support and propaganda. However, there has still not been an empirically oriented approach to understanding foreign support that focuses on support disaggregation. 

    Civil War Recurrence and Peace Duration

    According to a prominent article concerning foreign sponsors in civil war states (when summarizing the finding of previous work): “Civil wars with outside involvement typically last longer, cause more fatalities, and are more difficult to resolve through negotiations” (Salehyan, Gleditsch, and Cunningham 2011). Given that the literature has produced such unequivocal findings concerning the impact of foreign sponsors on civil war, it is surprising that the role of foreign sponsors in civil war recurrence has not been fully explored to a greater extent. 

    The literature on civil war recurrence has established a laundry list of factors that influence a return to conflict: the amount of time elapsed since the previous war, the outcome of the previous conflict, the presence of peacekeeping forces post-conflict, the duration of the previous war, the post-war size of the government military, the number of fatalities in the previous war, the level of economic development, the partition of the state, and proxies for the quality of life such as infant mortality (Collier and Hoeffler 2004; Quinn, Mason, and Gurses 2007; Walter 2004). This list reveals an important shortcoming in the civil war recurrence literature—the inadequate consideration of transnational factors (the only one in the above list is the presence of peacekeeping forces post-conflict).  This gap seems peculiar given the rich literature on foreign sponsors and war onset (discussed in the previous section). 

    Authors have identified conditions that impact the likelihood of the recurrence of civil war at multiple levels of analysis. Quinn, Mason, and Gurses (2007) conceptualize civil war recurrence as having two necessary conditions—one structural, one agency-based.  The necessary structural condition is dual sovereignty—which is present when a group other than the states possesses the capacity to challenge the state’s claim to sovereign authority. If the condition of dual sovereignty is not attained, there exists no effective opposition to the state, and thus, no civil war occurs. The necessary agency-based condition is that the expected utility of war is higher than the expected utility of peace for at least one of the potential combatants. If this condition is not met, then no actor would choose to reinitiate war (assuming a rational choice framework). Walter (2004) conceptualizes civil war recurrence as a product of individual-level decisions—specifically, whether or not to join a rebel organization—based on quality of life and the ability to affect politics though non-violent means. 

    Neither Quinn, Mason, and Gurses (2007), nor Walter (2004) explores the role of foreign sponsors in civil war recurrence. This ought to have occurred, given that foreign sponsors have the capability to affect conditions of structure (by reestablishing dual sovereignty by increasing the capacity of opposition groups) and conditions of agency (by reducing the costs of rebellion and altering the cost-benefit analysis of both combatant groups and individuals—the decision-making agents in the two models mentioned above). The purpose of this project is thus to address the aforementioned gaps in the literature by assessing the impact of foreign sponsorship on civil war recurrence.

    Theory and Hypotheses

    This project rests on the fundamental premise that cases of civil war recurrence are theoretically distinct from cases of first-time civil war onset. Simply coding cases of civil war recurrence as civil war onset misses the unique properties of recurring civil wars, many of which are grounded in the notion of the “the conflict trap” (Collier and Sambanis 2002; Collier et al. 2003). A country is caught in a conflict trap when a previous civil war has established a post-conflict environment that predisposes the country to experience a second civil war, which in turn will create another post-conflict environment that is ripe for renewed conflict. Collier et al. (2003) make a sound empirical case for the presence of the conflict trap—the best predictor of whether a country will find itself in civil war next year is whether it is experiencing civil war now, and 44 percent of countries experience a second civil war within five years of the initial war’s termination. It has been argued that cases where there is adequate motivation for civil war are fairly common (whether these motives are understood to be profit-driven or primordialist), and that the presence of civil war viability/capacity/opportunity often functions as a sufficient condition for civil war occurrence (Collier and Hoeffler 2004; Collier et al. 2003; Salehyan 2009).  The perspective of this project is that focusing on the capacity for civil war—as opposed to the motivation for civil war—is the better way to theorize about civil war recurrence. This project assumes that a previous civil war is a sufficient reason to assume the presence of adequate motivation for civil war. This means that if civil war is not recurring, it must be due to a lack of capacity to engage in warfare. This project will argue that foreign sponsorship plays a primary role in building combatants’ capacity for warfare. 

    There are two necessary conditions for the recurrence of civil war: an opposition group must be capable of challenging the state’s claim to exclusive sovereign authority (dual sovereignty); and a potential rebel group must be able to rationalize conflict reignition—the expected utility of war must be greater than the expected utility of peace for at least one would-be rebel organization (Quinn, Mason, and Gurses 2007). While there are many other possible necessary conditions for civil war recurrence, my model focuses in on these two.  Another possible necessary condition comes from Walter (2004): individuals must be able to rationalize engaging in rebellion—the expected utility of participating in rebellion must be greater than the expected utility of supporting the state or inaction. The presence of foreign sponsorship increases the probability that each of the necessary conditions is fulfilled. 

    For the re-establishment of dual sovereignty to occur, the opposition must become sufficiently powerful to challenge the state’s claim to sovereign authority. While many domestic factors can influence the strength of the opposition, foreign intervention can have dramatic effects on the relative capabilities of the combatants (Regan 2000).  Foreign support in the form of military equipment and foreign fighters in particular dramatically increases the ability of an opposition group to challenge the state. For war reignition to become rational for a potential combatant group, the expected utility of war must be greater than the expected utility of the status quo. Foreign support can change the expected utility of conflict by altering the probability of victory, the payoffs of victory and defeat, and the bargaining leverage of the warring parties. If foreign sponsorship is present, then there is an increased probability that the sovereign authority of the state is not absolute and that rational combatants would be more likely to initiate war. If there is an increased probability that those two conditions are met, then there is an increased probability that civil war recurs. This logic leads to this study’s first hypothesis:

    H1The presence of foreign sponsorship is associated with a shorter peace duration.

    H1 operates under the assumption that all types of foreign sponsorship increase the probability of civil war recurrence. This, however, may not necessarily be that case. It is possible that some forms of foreign support are primarily offensive or defensive in nature. The offense-defense balance literature pertaining to interstate war has acknowledged that some forms of material capability may be offensive or defensive in nature. When the scales are tipped in favor of offense, conquest is easier and security dilemmas result—encouraging self-interested states to pursue further armaments at the expense of others’ security, incentivizing others to do the same (Jervis 1978; Van Evera 1998). The roles that offensive and defensive technologies play in interstate war can be extended to offensive and defensive forms of external support provided to combatants in intrastate wars. Offensive forms of support—those that increase the capacity for accomplishing revisionist goals—could lead to a higher probability of recurrence. Increased capacity for changing the status quo is likely to drive actors who were previously dissatisfied, but incapable of affecting change, to attempt a revision of the status quo. Defensive types of support—those that increase the capacity for protecting the status quo power relationships—could lead to a lower probability of recurrence.  Such status quo entrenching capacity is likely to make altering the status quo more difficult.  Ambiguous types of support—those that increase the capacity for both offense and defense—should not significantly impact peace duration. These types of support are likely to contain countervailing tendencies (they are offensive in some way, defensive in others).  Specific examples of what this study considers offensive, defensive and ambiguous support are provided in the research design.These countervailing tendencies are likely to result in the support not substantially altering the expected utility of war relative to the expected utility of the status quo.  If this is the case, then H1 is not likely to be supported due to the opposing effects of offensive and defensive types of support—and the opposing effects of ambiguous types of support.  However, the following three-part hypothesis should be supported if distinct offensive and defensive forms of support have an impact on recurrence:

    H2AThe presence of offensive foreign support is associated with a shorter peace duration.

    H2BThe presence of defensive foreign support is associated with a longer peace duration.

    H2CThe presence of ambiguous foreign support has no significant effect on peace duration.

    If H2A and H2B are supported, and H1 is not supported, then there is strong support for conceptualizing types of foreign support as offensive or defensive.  Once support types are disaggregated by strategic function, disaggregation by the receiving party—the state or a rebel group receiving the aid—becomes theoretically useful.  If some types of support are not fungible across purposes—if they are primarily usable for distinctly offensive or defensive purposes, but not both—then actors with different strategic needs will appropriately assign different values to different forms of support. 

    If it is assumed that states merely wish to remain in power (preserve the status quo), then states have little use for offensive action because they have no logical reason to start a civil war.  Therefore, offensive support given to the state is not likely to enter into the decision-making process that determines whether rebels initiate war. Defensive support given to the state, however, is likely to make the state more difficult to dislodge and should push the cost-benefit analysis of rebel groups in the direction of not initiating war.  From this logic, a two-part hypothesis concerning support for states follows:

    H3AThe presence of defensive foreign sponsorship for the state is associated with a longer peace duration.

    H3BThe presence of offensive foreign sponsorship for the state has no significant effect on peace duration.

    If it is assumed that rebel groups wish first to survive (preserve an element of the status quo—their existence), and wish second to successfully defeat the state in war (revise an element of the status quo—the state’s claim to be the sole sovereign authority), then rebel groups value both offensive and defensive support—both of which are likely to alter the decision-calculus concerning war reignition.  The provision of offensive support increases the rebel group’s capacity to accomplish its revisionist goals, and therefore, the likelihood of the rebel group going to war. The provision of defensive support does not affect the rebel group’s capacity to challenge the state, but it does increase the chances that the rebel group will continue to exist as a political unit—allowing the rebels to “live to fight another day.”  Rebels are less likely to initiate a conflict when receiving defensive support since it would jeopardize one of their goals (survival), and it also does not increase their ability to accomplish their other goal (defeat the state).  From this logic, a two-part hypothesis concerning support for rebel groups follows:

    H4AThe presence of defensive foreign sponsorship for rebels is associated with a longer peace duration.

    H4BThe presence of offensive foreign sponsorship for rebels is associated with a shorter peace duration.

    Finally, some forms of foreign support that are not clearly offensive or defensive can be made to function primarily offensively or defensively by the type of actor that receives the support. The provision of foreign support via foreign troops is ambiguous in the hands of the state—the troops are offensive when used offensively, and defensive when used defensively (Regan 2002). For this reason (the inability to classify foreign troops as offensive or defensive), the presence of foreign troops in general, or those aiding a state specifically, is not expected to have any noticeable effect on the decision of rebels to engage in war. For rebel groups, however, foreign fighters serve primarily offensive, revisionist purposes due to the specific revisionist intentions inherent in rebel groups.  For this reason, foreign troop aid can be assumed to be offensive in nature when the aid is being given to a rebel group.  In other words, foreign troop aid is conditionally offensive (the condition for the offensive nature of the aid is that it must be given to a rebel group). From this logic, a three-part hypothesis follows:

    H5AThe presence of foreign fighters aiding rebels is associated with a shorter peace duration.

    H5BThe presence of foreign fighters aiding the state has no significant effect on peace duration.

    H5CThe presence of foreign fighters generally has no significant effect on peace duration.

    Research Design

    This section will discuss the plan to be implemented in testing the hypotheses presented above. Specifically this section will include the following: 1) identification of the sample (universe of cases); 2) operationalization of concepts key to the hypotheses being tested; and 3) discussion of the analytical technique utilized to test the hypotheses. 

    Sample Identification

    The temporal domain of this project is 1976 to 2009. This period has been selected primarily for methodological reasons: reliable data on the types of foreign support considered in this study are only available for this period of time. The UCDP External Support Dataset (Högbladh, Pettersson, and Themnér 2011) contains accurate data from 1975 to 2011. My operationalization of support provision (discussed in the next section) further limits my temporal domain by one year. While it would be preferable to consider all cases from 1945 to present, this domain is appropriate theoretically since the late 20th century saw a drastic increase in civil wars, many of which were recurring conflicts. The spatial domain of this project is global, encompassing all cases of civil war termination in the temporal domain. Including all cases of civil wars is advantageous for the following reasons: 1) there is no theoretical expectation that the dynamics of civil war recurrence are significantly different across cultures or geographic regions; and 2) a global scope allows for this study to draw more general conclusions about the effect of foreign sponsorship on peace duration.  The unit of analysis for this project is the post-war country-year—each post-war year is treated as a potential year for war to recur.   

    Operationalization

    The dependent variable in this project is civil war recurrence. For the purpose of this analysis, civil war is operationalized in a similar fashion as the term “armed conflict” in datasets from the Uppsala Conflict Data Program (UCDP): “a contested incompatibility that concerns government and/or territory where the use of armed force between two parties, of which at least one is the government of a state, results in at least 25 battle-related deaths” (Gleditsch et al. 2002).  This project’s definition of civil war alters the UCDP definition by raising the battle-related deaths threshold to 1000. This was done in order to capture recurrences of major fighting.    For example, if a country experiences intense civil war (1000 or more battle-related deaths) for a number of years, followed by minor armed conflict for an extended period (less than 1000, but more than 25, battle-related deaths), and then more intense civil war, the UCDP operationalization results in observation of civil war for the entire period.  UCDP’s definition results in coding many wars that died-out (a couple years of intense fighting, followed by many years of lower level conflict) as a decades-long civil war, thus obscuring the recurrence of major fighting, which is what this project is fundamentally seeking to understand.  Battle-related deaths data is derived from the UCDP PRIO Battle Deaths Dataset 1946 to 2008 (Lacina and Gleditsch 2005).  This dataset contains three measurements for annual battle fatalities: the best estimate, the low estimate, and the high estimate. For many conflict-years, a best estimate could not be determined. I use the best estimate when available.  When the best estimate is missing, this project uses the average of the high and low estimate.  It is assumed that this value is closer to the true value than the most extreme estimate in either direction.  Furthermore, UCDP applied its definition of armed conflict to specific conflict dyads, but this project simply applies the high-threshold definition to a given country for a given year. As this project uses the unit of country-year (instead of conflict-year), the battle-related deaths for concurrent conflicts are combined for a given year.  For example, if there are two conflicts in the same country during the same year, and each conflict inflicts 500 casualties, that year is coded as a civil war year (since it meets the 1000-battle-deaths-threshold).  In other words, for a conflict to be included in the dataset constructed for this study, at least one year of the conflict must have 1000 or more battle-related deaths. The civil war lasts for as many years as the 1000-battle-related-deaths-threshold is met.  There is one exception to this rule: the war between the United States and al-Qaeda. The increased casualty threshold and UCDP criteria are all met for this case, however, this conflict is fundamentally different from a civil war.  Thus, the conflict was removed from the dataset. An exception is made if the battle-related deaths count falls below 1000 for only one year, in which case that year is coded as a lull in the fighting, not a peace year. If the battle-related deaths count falls below 1000 for consecutive years, those years are coded as peace years. If the battle-related death count for a given year meets the threshold after two or more peace years, then that year is coded as the start of a new civil war—a recur year. Due to methodological restrictions concerning my independent variable of interest (elaborated on below), initial wars must end in 1975 or later for their recurrences to be acknowledged in my dataset.

    There are numerous independent variables of interest in this project.  The first is the existence of foreign sponsorship.  For this variable, the judgments made in the UCDP External Support Project—Primary Warring Party Dataset are used for the creation of the variable external_exists, which is a binary variable indicating the existence of “clearly established external support” (Högbladh, Pettersson, and Themnér 2011).  A significant limitation of the external support data is that it only contains observations for country-years containing a conflict that meets the UCDP definition of armed conflict. While Högbladh, Pettersson, and Themnér (2011) have observations for some years coded as peace years by this study (those that meet the UCDP battle-related deaths threshold, but not the 1000-battle-related-deaths-theshold), many more years are missing observations.  This project proxies the provision of support during these years with the perceived availability of support. The perceived availability of support is estimated based on the most recent year for which an observation is present. In other words, the value for the most recent observation is imputed for the missing observations until the next observation containing a value of support occurs. This estimation is based on the assumption that both rebels and states base their perceived ability to attract foreign support on their ability to attract such support in the most recent year of conflict. 

    The remaining explanatory variables used in this project are all sub-categories of existence of foreign support. All support variables are present in three forms—general, rebel, and state. General support is present when either rebels or the state is supported; rebel support is present when rebels are supported; state support is present when the state is supported.  In country-years containing conflict between a state and multiple rebel groups, only one rebel group must receive support for that observation to be coded as having rebel support. This coding decision was made to keep the interpretation of the binary variables consistent. The UCDP External Support Project—Primary Warring Party Dataset (Högbladh, Pettersson, and Themnér 2011) simply codes for the presence or lack of presence of support. Therefore, this study has simply coded for the presence of support for any rebel group—this project does not take into account which or how many rebel groups receive the support. While this project does not take into account which or how many rebel groups receive support, this is partially controlled for with the multiple rebel groups variable—which will be discussed below.

    This project uses a proxy for offensive support: “Weapons support” in the UCDP External Support Project—Primary Warring Party Dataset. Weapons support “includes donations, transfers, supplies, or loans of weapons or ammunition of any kind.” (Högbladh, Pettersson, and Themnér 2011).  This category of support does not perfectly conform to the ideal-type of offensive support, but it is the closest to the ideal-type of all forms in the dataset.

    A proxy is also used for defensive support: “Materiel/Logistics Support” in the UCDP External Support Project—Primary Warring Party Dataset. Materiel and logistical support includes many forms of non-weapon supplies.  Examples include vehicles, field hospitals, construction bulldozers, troop transportation, and repair and support facilities (Högbladh, Pettersson, and Themnér 2011). This category of support is too broad to be entirely defensive, but is the closest to a purely defensive support type.

    For this study, ambiguous support is also operationalized via proxy: “Funding/Economic Support” in the UCDP External Support Project—Primary Warring Party Dataset. Funding and economic support includes cash donations, military loans, and military grants, but does not include humanitarian, development, or balance of payments aid/loans (Högbladh, Pettersson, and Themnér 2011). This category is the best proxy for ambiguous support, since financial resources are the most fungible across offensive and defensive military objectives. 

    For the final independent variable of interest, rebel troop support is measured identically to Högbladh, Pettersson, and Themnér (2011). This type of support is included as a potential conditionally-offensive form of support (the expected condition for qualifying as offensive is that the troop support must go to rebels, not the state).  No type of support in the UCDP dataset can adequately explore the possibilities of conditionally-defensive forms of support, so none is operationalized here.

    Numerous variables must be controlled for in order to isolate the effect of foreign sponsorship of civil war recurrence. First, gross domestic product per capita (GDP per capita) is controlled for using the data from The Economic Statistics Branch of the United Nations Statistics Division (UNSD, 2012). This variable is used to estimate the wealth of a country.  Since a lack of wealth is associated with civil war, not including this control variable would bias the results of the analysis. Second, level of democracy is controlled for using the polity2 variable from the Polity IV Project (Marshall, Gurr and Jaggers 2013). Since different levels of democracy have been associated with civil conflict, the exclusion of this control variable would result in omitted variable bias. Both GDP per capita and polity2 are lagged by one year in this project’s models, because economic and political conditions from the previous year are more likely to affect civil war recurrence than similar observations for the year itself—observations for the year are could be highly distorted by the presence (or lack) of civil war.   Third, the existence of a previous peace agreement at the conclusion of the previous period of conflict is included as a control. Data for this variable was taken from the UCDP Conflict Termination Dataset (Kreutz 2010). This variable is controlled for because the presence of a peace agreement is thought to be negatively associated with the probability of civil war recurrence. Fourth, since this project required consolidating observations for concurrent conflicts, a control for multiple rebel groups was used in the analysis.  This variable was generated during the process of consolidating concurrent conflicts by country-year, and is controlled for because rebel infighting may affect the ability to reignite war.

    Analytical Technique

    Each model used in this project is a Cox proportional hazards model, which is utilized to measure the duration of peace. The “hazard”/failure used in the models is civil war recurrence—the recur year. Each model includes identical control variables: lagged GDP per capita, lagged polity2, previous peace agreement, and multiple rebel groups. The independent variable(s) of interest differ depending on the hypothesis being tested, but in each case consists of one or more types of support. For each model in the analyses, there were a total of 520 observations. In order to test H1, a model using the existence of foreign sponsorship and the aforementioned controls will be used. 

    For the test of H2, several models will be employed: three will include each of the categories of support (offensive support, defensive support, and ambiguous support) on their own; one includes the two types of support expected to have significant results (offensive support and defensive support); and one that includes all three categories in the same model. 

    In order to test H3, a model including offensive support for the state and defensive support for the state will be used. Similarly, the test for H4 will be a model including offensive support for rebel groups and defensive support for rebel groups

    Finally, to test H5, a series of models is utilized:  for each troop support recipient category (general troop support, troop support for rebel groups and troop support for the state), a model containing only that recipient category will be used; and another three models will pair each troop support recipient category with general offensive support and general defensive support

    Results and Analysis

    The analysis did not support H1 (The presence of foreign sponsorship is associated with a shorter peace duration). Model 1A shows that the existence of external sponsorship variable was not significant by any conventional standard. This is possible evidence that general support has no clear effect on civil war recurrence. Further evidence can be found in Model 1B (which finds existence of external sponsorship for the state to be insignificant), Model 1C (which finds existence of external sponsorship for rebel groups to be insignificant), and Model 1D (which finds existence of external sponsorship for the state and existence of external sponsorship for rebel groups insignificant when assessed together). See full results in Model 1B, Model 1C and Model 1D

    However, the analysis did support H2 (H2A):  The presence of offensive foreign support is associated with a shorter peace duration; H2B: The presence of defensive foreign support is associated with a longer peace duration; H2C: The presence of ambiguous foreign support has no significant effect on peace duration).  Model 2A found that, when assessed independently, offensive support is insignificant. Model 2B found that, when assessed independently, defensive support is significant at a 95 percent confidence level. Model 2C found that, when assessed independently, ambiguous support is insignificant. See full results in Model 2A, Model 2B and Model 2C.While Model 2B provides compelling results, Model 2D revealed a much more interesting finding: that when both offensive support and defensive support are included in a model, both variables are highly significant, have the expected impact on peace duration, and have large magnitudes. When paired together, both offensive support and defensive support are significant at a 99 percent confidence level.  Furthermore, the presence of offensive support increases the probability of civil war recurrence for a given year by 680 percent, while the presence of defensive support decreases the probability of civil war recurrence for a given year by 93 percent.  When ambiguous support is included in a model with offensive support and defensive support—as is the case with Model 2Eambiguous support remains insignificant. See  full results in Model 2E

    H3 (H3A):  The presence of defensive foreign sponsorship for the state is associated with a longer peace duration; H3B: The presence of offensive foreign sponsorship for the state has no significant effect on peace duration is partially supported by Model 3A, which includes both offensive support for the state and defensive support for the state.  H3A is supported at a 99 percent confidence level: defensive support for the state does decrease the probability of recurrence.  Furthermore, at a 98 percent decrease in the probability of civil war recurrence, the magnitude of the effect from defensive support for the state is greater than the magnitude of general defensive support—which is intuitive, given that a state’s ability to maintain itself as a political unit is fundamental to discouraging challenges to the state’s claim on sovereign authority. H3B, however, must be rejected, as offensive support for the state is significant at a 95 percent confidence level. While the theory underlying this project assumed that the state has little use for offensive capacity unless it is currently engaged in a civil war, there are plausible reasons for the state to value offensive capability—and for the presence of foreign support that boosts that capability to lead to renewed conflict.  For example, a state could use offensive capability to wipe out rebel groups with whom they have a cease-fire agreement in order to ensure that the rebel group will not have a chance to violate the agreement first.  A state could also use offensive capabilities to launch a campaign into a neighboring country that is providing safe haven to rebels that may threaten the state’s survival in the future.

    Both propositions of H4 (H4A):  The presence of defensive foreign sponsorship for rebels is associated with a longer peace duration; and H4B: The presence of offensive foreign sponsorship for rebels is associated with a shorter peace duration are supported by Model 4A.  While H4 is fully supported, some of the results do not fully reflect the theory presented in this paper. The variable offensive support for rebel groups is only significant at a 90 percent confidence level, and has a smaller effect on peace duration than does offensive support for the state. Furthermore, defensive support for rebel groups is less significant (at 95 percent confidence) than defensive support for the state. This could reflect less accurate data concerning support for rebels, but it could also reflect weaknesses in the theory proposed by this study.

    The final hypothesis—H5 (H5A): The presence of foreign fighters aiding rebels is associated with a shorter peace duration; H5B: The presence of foreign fighters aiding the state has no significant effect on peace duration; H5C: The presence of foreign fighters generally has no significant effect on peace duration—is fully supported by the analysis. Model 5A found that troop support for rebel groups is significant when tested independent of the other foreign fighter variables. See full results in Model 5A. A much more interesting finding comes from Model 5B, which includes not only troop support for rebel groups, but also offensive support and defensive support. All three variables have the expected effects on peace duration, all three are of a large magnitude, and all three are significant (two at 99 percent confidence and one at 95 percent confidence). The hazard ratio and significance level for offensive support were lower than those reported for Model 2D. The only difference between Model 2D and Model 5B is the inclusion of troop support for rebel groups.  This, along with the hazard ratios being in the same direction, implies that offensive support and troop support for rebel groups share explanatory power, which would be the case if troop support provided to rebels functions as an offensive form of support. The magnitude of troop support for rebel groups is also meaningful: the provision of troops to aid potential rebels increases the probability of civil war recurrence by nearly 800 percent. Model 5C and Model 5D assess the explanatory power of troop support for the state alone and paired with offensive support and defensive support. Model 5D and Model 5F do the same for general troop support—troop support for either rebel groups or the state. All four models return insignificant results for the troop support variables.  See full results in Model 5C, Model 5D, Model 5E and Model 5F.The high significance of troop support for rebel groups, paired with the lack of significance of troop support for the state and general troop support, implies that troop support is a conditionally offensive form of support. Troop support appears to be offensive in the hands of rebels, but ambiguous in the hands of the state and generally. 

    Policy Implications and Avenues for Future Research

    These findings have uncovered a series of salient policy implications:  the types of support outside actors provide can extend (or cut short) peace duration in post-conflict environments. By providing defensively oriented support—such as field hospitals, construction equipment, and maintenance of immobile defensive equipment—external parties can decrease the likelihood of war recurrence.  By providing offensively oriented support—even to actors dedicated to protecting the status quo—external parties will increase the probability of recurrence. Potential interveners thus cannot simply ask if intervention should happen and who they should support: they must also ask how they ought to support them. Furthermore, many outside parties try to foster stability by “propping-up” regimes with offensive support to be used for repressing dissent before it can transform into rebellion. The analysis done in this study shows that providing such offensive support is more likely to result in civil war than stability. 

    There is still much to be done to improve our understanding of how foreign sponsorship impacts civil war recurrence.  First, one of the clearest limitations of this project is the lack of accurate data on the availability of foreign support during peace years.  Collection of such data would allow for a more precise analysis that does not rely on assuming that potential combatants base their perceived ability to attract support on the support provided during the most recent war year. 

    Second, the differences in support recipients should be explored further. The weakest results in this project concerned H3 and H4 (the hypotheses addressing states and rebel groups separately). Furthermore, this project’s findings concerning conditionally offensive and conditionally defensive categories of support warrant a more nuanced theory on how states and rebel groups utilize external support differently. 

    Third, the variety of support providers should be explored. This project relied on data that only measured support from foreign states. Diasporas, refugee camps, foreign rebel organizations, criminal organizations, and even wealthy individuals can also support combatants, and may do so in fundamentally different ways than foreign states.  Some categories of sponsors may disproportionally provide offensive or defensive support, which would have significant implications for the theory presented in this study. 

    This paper set out to explore the role of foreign sponsorship on civil war recurrence, and has done so using duration analysis. Doing so has filled an important gap in the peace duration / civil war recurrence literature by considering external support as a primary catalyst for recurrence and disaggregating external support into three primary components: offensive support, defensive support, and ambiguous support. The results of testing the impact of offensive and defensive support yielded both high magnitudes and significance: not only do different types of support matter a great deal, but this study provides robust support for these findings.  Furthermore, the results regarding troop support (troop support for rebels is highly significant with a large magnitude, while both general troop support and troop support for the state are insignificant) suggest the possibility of conditionally offensive—and conditionally defensive—support types. 

    References

    • Adamson, Fiona B. 2013. "Mechanisms of diaspora mobilization and the transnationalization of civil war." In Transnational Dynamics of Civil War, ed. Jeffrey T. Checkel. New York, NY: Cambridge University Press.
    • Byman, Daniel, Peter Chalk, Bruce Hoffman, William Rosenau, and David Brannan. 2001. Trends in Outside Support for Insurgent Movements. Santa Monica, CA: RAND.
    • Collier, Paul, V. L. Elliot, Havard Hegre, Anke Hoeffler, Marta Reynal-Querol, and Nicholas Sambanis. 2003. Breaking the Conflict Trap. Washington, DC: The International Bank for Reconstruction and Development / The World Bank.
    • Collier, Paul, and Anke Hoeffler. 2004. “Greed and Grievance in Civil War.” Oxford Economic Papers 56(4): 563–95.
    • Collier, Paul, and Nicholas Sambanis. 2002. “Understanding Civil War: A New Agenda.” The Journal of Conflict Resolution 46(1): 3–12.
    • Fearon, James D., and David D. Laitin. 2003. “Ethnicity, Insurgency, and Civil War.” American Political Science Review 97(01): 75–90.
    • Gleditsch, Kristian Skrede. 2007. “Transnational Dimensions of Civil War.” Journal of Peace Research 44(3): 293–309.
    • 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): 615–637.
    • Högbladh, Stina, Therése Pettersson, and Lotta Themnér. 2011. "External Support in Armed Conflict 1975-2009 -presenting new data." Unpublished manuscript presented at the International Studies Association Convention in Montreal 2011.
    • Jervis, Robert. 1978. "Cooperation under the security dilemma." World Politics 30(2): 167-214.
    • Kreutz, Joakim. 2010. "How and When Armed Conflict End: Introducing the UCDP Conflict Termination Dataset." Journal of Peace Research 47(2): 243-250.
    • Lacina, Bethany, and Nils Petter Gleditsch. 2005. "Monitoring Trends in Global Combat: A New Dataset of Battle Deaths." European Journal of Population 21(2-3): 145-116.
    • Marshall, Monty G., Ted Robert Gurr, and Keith Jaggers. 2013. Polity IV Project: Political Regime Characteristics and Transitions, 1800-2012. Center for Systemic Peace.
    • Quinn, J. Michael, T. David Mason, and Mehmet Gurses. 2007. “Sustaining the Peace: Determinants of Civil War Recurrence.” International Interactions 33(2): 167–93.
    • Regan, Patrick M. 2000. Civil Wars and Foreign Powers: Outside Intervention in Intrastate Conflict. Ann Arbor, MI: The University of Michigan Press.
    • Regan, Patrick M. 2002. "Third-party interventions and the duration of intrastate conflicts." Journal of Conflict Resolution 46(1): 55-73.
    • Salehyan, Idean. 2008. “No Shelter Here: Rebel Sanctuaries and International Conflict.” The Journal of Politics 70(1): 54–66.
    • Salehyan, Idean. 2009. Rebels Without Borders: Transnational Insurgencies in World Politics. Ithaca, NY: Cornell University Press.
    • Salehyan, Idean, Kristian Skrede Gleditsch, and David E. Cunningham. 2011. “Explaining External Support for Insurgent Groups.” International Organization 65(Fall 2011): 709–44.
    • United Nations Statistics Division, Economic Statistics Branch. 2012. "National Accounts Estimates of Main Aggregates." http://data.un.org/Data.aspx?q=GDP+per+capita&d=SNAAMA&f=grID%3a101%3bcurrID%3aUSD%3bpcFlag%3a1#SNAAMA (July 22, 2013).
    • Van Evera, Stephen. 1998. "Offense, Defense, and the Causes of War." International Security 22(4): 5-43.
    • Walter, Barbara F. 2004. “Does Conflict Beget Conflict ? Explaining Recurring Civil War.” Journal of Peace Research 41(3): 371–88.

    Table 1: Model 1A

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    existence of foreign sponsorship

    0.536

    0.299

    0.264

    lagged GDP per capita

    0.999

    0.001

    0.052*

    lagged polity2 score

    0.899

    0.050

    0.055*

    previous peace agreement

    0.609

    0.392

    0.441

    multiple rebel groups

    0.551

    0.431

    0.446

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 2: Model 1B

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    foreign sponsorship for the state

    0.477

    0.261

    0.176

    lagged GDP per capita

    0.999

    0.001

    0.060*

    lagged polity2 score

    0.909

    0.048

    0.072*

    previous peace agreement

    0.463

    0.293

    0.224

    multiple rebel groups

    0.416

    0.335

    0.276

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 3: Model 1C

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    foreign sponsorship for rebel groups

    1.479

    0.844

    0.493

    lagged GDP per capita

    0.999

    0.001

    0.051*

    lagged polity2 score

    0.906

    0.049

    0.067*

    previous peace agreement

    0.475

    0.315

    0.261

    multiple rebel groups

    0.434

    0.359

    0.313

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 4: Model 1D

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    foreign sponsorship for the state

    0.430

    0.243

    0.136

    foreign sponsorship for rebel groups

    1.750

    1.048

    0.350

    lagged GDP per capita

    0.999

    0.001

    0.061*

    lagged polity2 score

    0.910

    0.048

    0.071*

    previous peace agreement

    0.358

    0.249

    0.140

    multiple rebel groups

    0.316

    0.278

    0.191

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 5: Model 2A

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    offensive support

    1.618

    1.571

    0.620

    lagged GDP per capita

    0.999

    0.001

    0.079*

    lagged polity2 score

    0.923

    0.084

    0.401

    previous peace agreement

    0.423

    0.444

    0.413

    multiple rebel groups

    0.435

    0.420

    0.389

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 6: Model 2B

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    defensive support

    0.275

    0.154

    0.021**

    lagged GDP per capita

    0.999

    0.001

    0.057*

    lagged polity2 score

    0.869

    0.055

    0.026**

    previous peace agreement

    0.738

    0.463

    0.628

    multiple rebel groups

    0.629

    0.470

    0.535

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 7: Model 2C

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    ambiguous support

    1.127

    0.787

    0.864

    lagged GDP per capita

    0.999

    0.001

    0.064*

    lagged polity2 score

    0.908

    0.061

    0.148

    previous peace agreement

    0.535

    0.392

    0.394

    multiple rebel groups

    0.475

    0.486

    0.467

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 8: Model 2D

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    offensive support

    7.796

    6.101

    0.009***

    defensive support

    0.072

    0.055

    0.001***

    lagged GDP per capita

    0.999

    0.001

    0.079*

    lagged polity2 score

    0.886

    0.064

    0.094*

    previous peace agreement

    0.449

    0.504

    0.317

    multiple rebel groups

    0.590

    0.359

    0.537

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 9: Model 2E

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    offensive support

    7.289

    6.936

    0.037**

    defensive support

    0.071

    0.053

    0.000***

    ambiguous support

    1.189

    0.896

    0.818

    lagged GDP per capita

    0.999

    0.001

    0.081*

    lagged polity2 score

    0.883

    0.065

    0.093*

    previous peace agreement

    0.447

    0.370

    0.331

    multiple rebel groups

    0.560

    0.565

    0.566

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 10: Model 3A

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    offensive support for the state

    7.635

    7.697

    0.044**

    defensive support for the state

    0.020

    0.030

    0.007***

    lagged GDP per capita

    0.999

    0.001

    0.107

    lagged polity2 score

    0.872

    0.055

    0.030**

    previous peace agreement

    0.351

    0.267

    0.168

    multiple rebel groups

    0.341

    0.319

    0.250

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 11: Model 4A

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    offensive support for rebel groups

    6.163

    6.043

    0.064*

    defensive support for rebel groups

    0.141

    0.128

    0.031**

    lagged GDP per capita

    0.999

    0.001

    0.084*

    lagged polity2 score

    0.903

    0.056

    0.152

    previous peace agreement

    0.491

    0.415

    0.400

    multiple rebel groups

    0.566

    0.504

    0.523

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 12: Model 5A

     

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    troop support for rebel groups

    6.960

    6.593

    0.041**

    lagged GDP per capita

    0.999

    0.001

    0.062*

    lagged polity2 score

    0.907

    0.068

    0.192

    previous peace agreement

    0.473

    0.400

    0.376

    multiple rebel groups

    0.537

    0.441

    0.449

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 13: Model 5B

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    offensive support

    6.347

    5.101

    0.021**

    defensive support

    0.062

    0.047

    0.000***

    troop support for rebel groups

    8.984

    7.646

    0.010***

    lagged GDP per capita

    0.999

    0.001

    0.045**

    lagged polity2 score

    0.882

    0.066

    0.093*

    previous peace agreement

    0.470

    0.388

    0.361

    multiple rebel groups

    0.716

    0.593

    0.687

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 14: Model 5C

     

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    troop support for the state

    0.976

    0.754

    0.975

    lagged GDP per capita

    0.999

    0.001

    0.053*

    lagged polity2 score

    0.909

    0.063

    0.167

    previous peace agreement

    0.540

    0.392

    0.395

    multiple rebel groups

    0.504

    0.402

    0.390

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 15: Model 5D

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    offensive support

    7.684

    6.199

    0.011**

    defensive support

    0.071

    0.055

    0.001***

    troop support for the state

    0.872

    0.664

    0.857

    lagged GDP per capita

    0.999

    0.001

    0.080*

    lagged polity2 score

    0.887

    0.065

    0.100*

    previous peace agreement

    0.442

    0.352

    0.305

    multiple rebel groups

    0.591

    0.504

    0.537

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 16: Model 5E

     

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    troop support

    2.009

    1.421

    0.324

    lagged GDP per capita

    0.999

    0.001

    0.069*

    lagged polity2 score

    0.895

    0.073

    0.175

    previous peace agreement

    0.584

    0.439

    0.474

    multiple rebel groups

    0.481

    0.393

    0.371

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010

    Table 17: Model 5F

    _t

    Hazard Ratio

    Robust SE

    P>|z|

    offensive support

    8.243

    6.558

    0.008***

    defensive support

    0.071

    0.051

    0.000***

    troop support

    1.886

    1.261

    0.343

    lagged GDP per capita

    0.999

    0.001

    0.088*

    lagged polity2 score

    0.879

    0.070

    0.106

    previous peace agreement

    0.499

    0.413

    0.401

    multiple rebel groups

    0.602

    0.512

    0.550

    Number of observations: 520; Number of failures: 15

    *significance level at p≤0.100    **significance level at p≤0.050    ***significance level at p≤0.010