Abstract
Previous research has uncovered only ambiguous evidence of the mechanisms that support or inhibit democratic trajectories in the aftermath of civil war. Here I suggest that one specific form of transnational aid during a civil war may have reverberating consequences after the fighting stops. Specifically, when a state emerges to control the executive after a conflict with the help of a previous interstate enemy, the leadership is vulnerable to political attacks on their patriotism and judgment. As such, open democracy becomes a less attractive option for these executives. I investigate this proposition using difference-in-difference matching estimation, as well as several alternative specifications. The findings strongly suggest the presence of disincentives to democratize for those executives that received help from external rivals. This research provides a new set of tools for identifying the causes and potential remedies to deficient democracy after civil wars.
Keywords
Introduction
Researchers have noticed that civil wars have produced an incomplete but intriguing record of democratization (Wantchekon, 2004; Fortna, 2009; Fortna & Huang, 2012). Peru and Thailand are examples of states that became more democratic despite the recent outbreak of a civil war. Alternative post-conflict paths are not difficult to identify however, including Uganda, Iran, Somalia, and Sri Lanka. In Wantchekon’s (2004: 17) tally, between 1946 and 1993, 40% of civil wars resulted in subsequent democratization. The top plot in Figure 1 illustrates snapshots of the level of democracy 1 at the end of the civil war (to the left) and ten years later (to the right). The thickness and the darkness of the lines are proportional to the scale of the absolute change. Dotted lines show decreases during this time. Here we see significant heterogeneity in post-conflict democratic trajectories; some states rise, some fall, others are relatively stable at different levels. The bottom plot in Figure 1 clarifies another facet of the postwar democratization puzzle. The boxplots summarize the annual distribution of changes, as compared to the levels, in post-conflict democracy scores since the end of the civil war. The underlying observations are plotted as points and the x’s represent the means. The pattern is one of relatively small movement in central tendency towards democracy, but with an increasing spread as time moves on. This suggests that there is indeed variance in post-conflict trajectories that calls for an explanation.
Several plausible candidates to explain this heterogeneity in post-conflict democratization have been identified, including the role of the interventions Top plot: the dots represent democracy levels at the end of the war and ten years after a civil war. The lines illustrate the linear trajectories. The line thickness and darkness represent changes of greater magnitude in absolute value. The dotted lines represent declines.
I argue that the choices made by rebel and government groups during a civil war to accept transnational aid from an external rival reduce the putative popular support that group would garner in a future free and open democratic election. While the importance of transnational interactions has been discussed within a conflict setting (Salehyan, 2009; Craft & Smaldone, 2002; Gleditsch, 2007), their influence post-conflict has not been systematically analyzed. When an executive in power is associated with a group that previously accepted aid from an international enemy (Akcinaroglu & Radziszewski, 2005), they have a disincentive to democratize, since their expected electoral support will be reduced. If the rival-provided aid is common knowledge, this can be used by opposition parties to paint the winner as perfidious rather than patriotic. Similarly, even if the support was not fully known during the conflict, free and fair elections raise the probability that the tainted aid would be revealed publicly. Therefore, a group that emerges from a civil war in a position of power, having been aided by an enemy state, has both the incentive and the opportunity to play the part of democratic spoiler. Using newly available data on international support for rebel groups and governments in civil wars, empirical tests using difference-in-difference matching methods provide strong support for the expectation that post-conflict democracy levels are lower when the emerging executive received aid from a external state rival.
The postwar democratization puzzle
Several research projects, including Wantchekon (2004), Fortna (2009), Fortna & Huang (2012), and Madhav (2010), attempt to identify the observable indicators that would successfully explain which states democratize. Wantchekon (2004) argues that former rebel groups and local leaders may use democracy to enhance their credibility in enforcing a political bargain. This analysis and the related model in Wantchekon & Neeman (2002) suggest that short-term external intervention may increase the probability of post-conflict democratization, as well as the presence of a heterogeneous electorate, whereby at least two competing parties in the civil war have a chance to win the subsequent election. Madhav (2010) plausibly suggests that postwar democratization occurs when a balance of power between rebels and the governments deters a renewal of fighting. Yet, despite these theoretical contributions, the empirical analysis of postwar democratization has continued to produce ambiguous findings (Fortna & Huang, 2012). While Gurses & Mason (2008) report evidence that a military victory on one side decreases the probability of postwar democratization, Madhav’s (2010) research finds no evidence that victory has any appreciable effect on post-conflict democratization. The estimated effect of ethnic heterogeneity on postwar democratization has been inconsistent, as has the role of the United Nations specifically in democratization (Doyle & Sambanis, 2000; Fortna, 2009).
There are myriad reasons for these disjointed inferences. Fortna & Huang (2012) highlight that this constellation of analyses uses different measurements, distinct time domains, and divergent specifications. However, the lack of robustness across research designs is at a minimum inconvenient to pre-existing theories linking civil war events to subsequent institutional changes.
Disaggregating the problem
One organizing cleavage in this cacophony of findings is a debate concerning whether postwar democratization is fundamentally different than democratization outside of a civil war context. For example, Madhav (2010) proposes that the previous civil war creates a drag on democratization due to the contentious relationship between elites and their fear of being victims of renewed violence. Alternatively, Fortna & Huang (2012) argue that, after weighing the available evidence, post-civil war democratization follows the same systematic patterns found in peaceful contexts. They conclude that their research ‘find[s] little support for the prominent notions that the war’s outcome, its death toll, or the presence of peacekeepers affects the prospects for democracy’ (Fortna & Huang, 2012: 801). Further, they infer that ‘democratization in societies emerging from civil war is shaped by much the same factors thought to affect democratization in other [non-civil war] societies’. To a large extent, war duration, intensity, and outcomes have been estimated to play at most a limited role in predicting which countries will democratize in the aftermath of civil war.
This need not be the last word on the linkages between civil war interactions and post-conflict context. It still could be the case that the key characteristics of civil wars that predict future democratization (or lack thereof) have not yet been tested. Specifically, it is of note that the empirical measures used to track democratization have all been aggregated over combatants’ choices, yielding variables such as duration, presence of a decisive outcome, and dyadic intensity, that abstract away from the specific decisions and concomitant transmission of information from rebel or government actors to civilians during a conflict. For example, duration is an aggregation of choices – while a rebel group could give up, leading to a government victory and a shorter duration, the government could simultaneously expand their demands, leading to continued fighting.
2
Similarly, the intensity of a war is a function of how hard both sides are trying, and whether one is hiding or not. In fact, Fortna & Huang (2012: 807) argue that, the burgeoning research on post-civil war democratization is in need of deeper theorizing about the incentives of the actors – the state, rebel groups, the domestic population, external actors – that are involved in making choices about postwar regimes. Given that a deadly civil war has occurred, who gains or loses with political liberalization, and why? Given the salience of the question of post-civil war democratization in current affairs, further scholarly research on the topic is clearly worthwhile.
It is this gap in our understanding of actors’ incentives to liberalize, clearly identified by Fortna & Huang (2012), that I attempt to partially fill here. I build on the important foundation set by Wantchekon (2004) and Madhav (2010) about the winners and losers of putative political democratization by analyzing the postwar repercussions of particular choices made by civil war actors during the conflict. One crucial transnational choice that has not received attention in the post-conflict democratization literature to date, despite being relatively common in civil conflict settings, has the potential to create anti-democratic incentives for future leaders. Specifically, if the leadership of a conflict group that emerges from the war with significant political power has previously accepted aid from an international enemy and rival, this makes it electorally vulnerable in any future election.
Transnational aid and civil conflict: Beware rivals bearing gifts
Gleditsch (2007) and others forcefully make the case that civil wars are not fought purely inside their own borders. Weapons are manufactured and smuggled from abroad, rebellions are housed across borders (Salehyan, 2009), and international forces are mobilized in response to and sometimes engage within the civil conflict (Doyle & Sambanis, 2000; Pickering & Peceny, 2006). The most intensely researched aspect of the internationalization of state–rebel conflict is its effect on escalation and settlement. External actors can increase the duration and intensity of civil conflict and can significantly aid the chances of rebel victory (Gleditsch & Beardsley, 2004; Gleditsch, 2007; Walter, 1997; Salehyan, 2009).
Additionally, Hegre & Sambanis (2006) find that civil wars often diffuse across borders, either through direct intervention or indirect conflict contagion related to increased supplies of arms. This has led others to explore the role civil conflict plays as a potential generator of state-to-state international conflict. Schultz (2010) reports evidence that external support for rebel groups can increase international conflict. For example, if a rebel group is in civil conflict with the government of state A, and external state B provides help to the rebel group, this dramatically increases the probability of interstate conflict between states A and B. Versions of this story have played out between Israel and its neighbors, as well as between Somalia and Ethiopia in the 1980s. 3
To the extent that international intervention has been included in studies of post-conflict democratization, the focus has been on democracy promoting mechanisms. 4 Influential research by Doyle & Sambanis (2000) and Fortna (2008) explicitly analyzes the role the UN plays in post-conflict settings. Pickering & Peceny (2006) likewise analyze the international role of liberal democratic states such as the USA, France, and the UK in post-conflict democratization. These studies suggest that international intervention has the goal of promoting democracy in the target state. However, the findings on the effectiveness of UN intervention on post-conflict democratization have been ambiguous (Fortna, 2009). 5
The supply side
What has been relatively neglected so far is the role of non-liberal motives in transnational civil settings. Akcinaroglu & Radziszewski (2005), Gleditsch & Gartzke (2005), and Colaresi (2006) explicitly make the case that states often have intervention motives that do not involve democracy (see also Bueno de Mesquita & Downs, 2006; Maoz & San-Akca, 2012). These can range from protecting ethnic kin, as in Albanian support for the KLA in Serbia, to stemming a refugee problem, as in Indian intervention in the East Pakistan civil war, to exploiting civil conflict to improve its international position, for example in Ethiopian support for Somali rebel groups after the Ogaden war or in Costa Rican support for the nascent FSLN.
A government involved in some level of violent domestic political competition that receives aid from an international ally with similar security concerns and preferences, as exemplified by US–Israeli, US–Spanish or North Korean–Chinese cooperation, would seem uncontroversial to the vast majority of the public in each country. Aid from the USA and the UK to the French resistance during World War II was not seen as controversial by the French public and did not inhibit resistance leaders such as De Gaulle from ascending to power and holding elections.
As Gowa (1995) has argued with respect to state-to-state economic trade, if a country has interests that significantly overlap those of another state, then strengthening that ally also strengthens its own international position. By the same logic, a state actor that has security interests in opposition to a given government in an external state may protect, support, and supply groups that compete for sovereign power within that external rival state, since that support helps undermine the source of an international security threat. These types of international and transnational sponsorships of external civil opposition are not new. During the run-up to World War II, for example, elites in both the United Kingdom and the United States rang the alarm about potential ‘fifth column’ support, and the warning was renewed at the beginning of the Cold War (Steele, 1999: 75; Paterson 1988: 10). Akcinaroglu & Radziszewski (2005) and Dreyer (2010) identify a significant proportion of international rivalries that involve one state supporting rebels against the other state. Thus, groups competing for power within a state provide a useful lever for external rivals to distract and weaken an international enemy. 6
The demand side
On the other side of the rival–rebel transaction, 7 the recipients of rival aid may view the aid in similarly Kauytilian terms – the enemy of my enemy is my friend. Despite being non-state actors, rebel groups often have coherent aims in contesting their state governments (Thomas, 2012). Similarly, a careful study by Lemke (2008) illustrates that even non-state actors often use short-term alliances based on common security interests to survive local threats. Thus, it should not be a surprise that rebel group leaders will be receptive to offers of support from potential arms donors.
Aid from a rival, in the immediate term, would be difficult to turn down if it helped to decrease the probability of defeat. In the conflict in the Democratic Republic of Congo (DRC), a group led by rebel general Bosco Ntaganda has coordinated and allied not only with the M23 rebel group, but international actors who share their common enemy – the Congolese government. Specifically, offers by Rwanda and Uganda of weapons and cover to the rebels have been accepted. A Human Rights Watch (2012) report states that international support for ‘Ntaganda’s forces by Rwandan military officials included Kalashnikov assault rifles, grenades, machine guns, and anti-aircraft artillery’ and that these weapons were viewed as highly valuable by the rebel group according to defectors that were interviewed. Systematic evidence suggests that rebel groups benefit from this international support. Having a base across the border from a civil conflict zone has a statistically significant and positive impact on a rebel group’s ability to continue contesting with a government (Salehyan, 2007).
Evidence of the market for rival-to-rebel support
Cases of an external state rival supporting one side in a civil war are not difficult to identify. Nicaragua’s rival Costa Rica 8 provided consistent and early help to the FSLN. Pastor (2002: 124–125) reports that Costa Rica supported the FSLN as a strategic choice to undercut its rival, and saw the FSLN as a useful tool to quell the Nicaraguan security threat. Costa Rica not only provided a base for escaped FSLN leaders such as Robelo and Ramirez, but also provided extensive logistical support and supplies that originated in Venezuela and Cuba (Pastor, 2002: 125–126).
Similarly, FROLINAT in Chad received substantial aid from Chad’s rival, Libya, which had contested the Aouzou strip. Libyan support in the fight was both direct and indirect, with FROLINAT headquartered in Tripoli and the Libyan military providing direct support at crucial junctures. Libya’s strategy at the time is described by Burr & Collins (1999: 226–227): ‘Qaddafi, however, could exploit any power vacuum or discontent in the periphery. For most of this period, disaffected politicians in N’Djamena were almost assured Libyan money and arms once they declared themselves as rebel leaders’. Other examples are the Khmer Rouge in Cambodia receiving support from Vietnam (Chandler, 1992), the Somali National Movement being provided bases in Ethiopia as well as air cover by the Ethiopian air force in 1982 (Gilkes, 1998), the Polisario receiving support from Algeria for their conflict with the Moroccan government (Arieff, 2012), and UNITA in Angola benefiting from aid donated by then-Zaire (Beit-Hallahmi, 1988: 65). 9
The consequences for post-conflict democratization
The presence of transnational aid from an enemy state will not only have consequences during the civil war, in the form of enduring conflagrations and rebel resilience as shown by previous research, but also will alter the incentives for democratization in the post-conflict political landscape. Specifically, this aid reduces the likely public support leaders could expect if they choose to democratize. For example, in Somalia during the three decades following independence, Ethiopia was an increasingly implacable rival that maintained control over the Ogaden which was claimed by Somalia. In fact, one of the points on the five-pointed star on the Somali flag referred to the Ogaden land that was claimed by Mogadishu. Several Somali dissident groups over the years have been explicitly supported by Ethiopia, including the SNM, and later the SRRC and the USC. However, those groups’ popular support, outside their immediate clan base, has consistently been compromised by their association with Ethiopia. The SNM were branded traitors and more recently the Transitional Government has been facing a legitimacy crisis because it is looked upon as being an Ethiopian puppet (Menkhaus, 2007; see also New York Times, 2008; Mail & Guardian, 2006). This occurred because Ethiopia, as a rival, was viewed as a public security threat to Somali interests. 10
Similar patterns were apparent in Nicaragua and Libya. Realizing the unpopularity of the external aid received during the crisis, and how it challenged their image as patriotic liberators, the FSLN made significant efforts to distance itself from Costa Rica after the conflict and was wary of post-conflict elections and censored press coverage of dissidents because it feared the putative opposition. The FSLN also delayed elections and reduced the openness of the competition (Pastor, 2002: 182–183; Vaden & Prevost, 1993). Libyan support for FROLINAT was highly controversial in Chad and even led to a split in the rebel leadership. The close relationship between Goukouni and Libya was extremely unpopular in Chad, given the historical competition over the Aouzou strip and Libya’s recent occupation of that disputed territory. Libyan support significantly reduced the popularity of the FROLINAT leadership, and in turn the chances of opening the political process to competition declined further (Azevedo, 1998: 147). In these cases, the civil war ties to an external rival were viewed as significant political baggage in the post-conflict context that reduced the prospects of open political competition in the post-conflict setting.
Public reactions
This public skepticism of a leader or group that survived a civil war through aid from a rival is rooted in two potential doubts, one prospective, the other retrospective. First, the private preferences of leaders are unobservable; therefore, citizens and other elites use the information available to them to forecast likely policies. 11 Does the leader value the national interest of the state, or is the leader a puppet for the rival’s interests? In one extreme example, it would be difficult to defend, on national security grounds, Goukouni’s announcement that he was going to unite Chad with Libya, after previously acquiescing to the Libyan annexation of the disputed Aouzou strip. In fact, the decision to ally so closely with Libya led even those within his leadership circle to question Goukouni’s patriotism (Azevedo, 1998). Therefore, aligning with a rival suggests future policies in the same placative direction, particularly to those opposition partisans that are not disposed to trust the potentially tainted leader.
Second, the public, if they were to learn of the rival aid, may electorally punish a leader for the perceived damage already done to national security. The choice to ally with a rival has the potential to detrimentally affect national security in several ways. Cooperating with an external rival may alert the rival to existing security vulnerabilities, such as routes for smuggling that could also be used for a militarized advance or provide intelligence on the location of state bases and other defenses. Similarly, if arms were procured from the rival, resupplying the military might increase dependence on that rival, strengthening the rival’s relative bargaining position. In Somalia, Ethiopia has benefited significantly from having a weakened Somalia, particularly relative to the tensions that led to the 1977 Ogaden War between the two countries. Opposition groups have stated that accepting Ethiopian aid has weakened Somali interests ( Mail & Guardian, 2006). Citizens may wish to punish, ex post, rather than reward leaders that ally with rivals, if they were given the chance at the ballot box. Of course, the public may not get that chance if the leader knows s/he is unpopular and rules through a smaller, as opposed to a larger winning coalition.
Due to these reasons, support from a rival, if it is known, will be called into question by the opposition, directly impugning the motives, judgment, and record of the group that accepts the aid, during a subsequent open debate. While there may be private benefits to the groups accepting the help, there are potential public costs. Opposition groups are likely, in any putative free and open election, to make the case that taking help from an external rival has signaled that the rebel or government side is willing to trade future national security for short-term specific gains.
The incentives and opportunities to spoil democratization
There is little reason, especially for a group that has emerged politically powerful from a civil war, to hold open, fair and free democratic elections that it is certain to lose. As Wantchekon (2004: 28) notes, post-civil war democratization will only be chosen when the side or sides that remain standing have a high enough probability of winning the subsequent election. Jarstad (2009: 371–374) adds that ‘[r]ebels are not always interested and able to transform into political parties, because if they emerge as a political party, they risk losing the election’. In some cases autocracy and repression are embraced by ruling elites, as in Chad. More subtly, Vaden & Prevost (1993: 55–56) specifically charge the FSLN in Nicaragua with avoiding free and fair elections after the end of the civil war because they were afraid of losing. If an actor in an important position of power has a lower expected utility for democratization, we should be less likely to see democratization, ceteris paribus.
In this way, looking at the external donors to the groups during a civil war should help predict post-conflict democratization trajectories, if those groups are in a position to impede democratization. Rebel groups or government forces that allied with external rivals
12
face the possibility of a post-conflict crisis of legitimacy in the eyes of the public. This leads to the primary hypothesis explored in this article:
Hypothesis: If an executive within a state is associated with a conlict group that received military aid from a rival state during the conflict, the level of post-conflict democratization will be lower than comparable cases where aid from a rival was not received.
This hypothesis suggests that it is the joint condition of a conflict-group accepting aid from a rival, while also being in a position of post-conflict political power that will forecast lower democracy trajectories. If a state accepts aid from a rival but definitively loses the conflict and is out of power, it may have the same incentive to stop elections, but a much reduced opportunity to effect institutional change. Alternatively, it could be the case that any aid from a rival into a civil conflict, on either side, poisons the well of democracy and legitimacy after the fighting. To explore this possibility, I will also test for the possibility that aid from rivals to losing sides in a civil war decreases democratization, ex post.
Clandestined to autocracy
It is important to note that this prediction, linking rival aid to post-conflict democratization, does not depend on whether the public, during the civil war, is aware of the source of military aid. Of course if the public is aware of the aid, the logic of the argument remains clear. However, what might be less obvious is that the disincentive to democratize still exists if the public is unaware of the previous cooperation with the rival. This occurs because, if the leader were to support transparency, open elections, and democratization, the rival aid would become publicly known as part of the electoral process and the expected utility of the democratic choice for the leader would be significantly reduced a priori. Some states work very hard to conceal their transnational support, most likely to avoid direct retribution from other states (Schultz, 2010). However, just as battles reveal the previously private strength and motivations of each side, elections, press freedom, and other accoutrements of democracy are likely to reveal controversial historical ties during the previous civil war. This potential for revelation decreases the incentives rival-aided executives have to democratize after the conflict. 13
Research design
Measuring rival support
To test whether there is empirical support for this hypothesis, I collect data on which civil war actors choose to accept aid from an external rival state. I use the Doyle & Sambanis (2000) list of civil wars covering the years 1946 to 1999. 14 I first code whether any of the actors in the civil war received material support from an external rival. Both the actors and their sources of international aid were coded from the Non-State Actor Dataset (NSA). 15 International rivals were coded using both the strategic rivalry and enduring rivalry (Diehl & Goertz, 2000) definitions from Colaresi, Thompson & Rasler (2007: 37–64). While the strategic rivalry definition defines an interstate rivalry as being one that involves the perception of enmity and threat by the executive, the enduring rivalry definition codes the density of militarized disputes between two countries. Both are likely to be relevant and in this set of cases there is substantial overlap. 16 The dyadic rivalry had to be coded as ongoing before the civil conflict started in order to be counted. I made one additional change to this rivalry coding. If the issues underlying the interstate rivalry were solely about support for rebel movements and not a larger national security issue, then I did not code the aid as coming from a rival. Keeping these causes would have risked mis-measuring policy disagreements between states. If the only rivalry issue on record was rebel support, this would not reflect underlying national security disagreements. Information on the issues underlying strategic rivalries comes from research by Colaresi, Rasler & Thomson (2007) and Dreyer (2010). 17
Coding the opportunity to spoil
After measuring whether each actor received material aid from a rival, I coded which actors in the dispute had representatives heading the executive after the civil war. This is similar to coding civil war winners in many cases, but not all. For example, inconclusive civil conflict outcomes, as in Iraq after the 2003 interstate conflict, can still place a new leader in power. I use links to the executive to code the opportunity to block democratization because this elite will be in the strongest position to steer the future institutional trajectory of the state or at the very least apply the brakes. While various measures of outcomes are available, we are interested in whether the executive of the country has an interest in democratization or is seen as tainted by the previous choices in the civil war. Leadership of the executive is coded from Goemans, Gleditsch & Chiozza (2009) and links to civil war conflict groups were identified through news and historical searches. 18
Measuring democracy
I code the democratic trajectory of each state after the end of a civil war using the xPolity data as coded by Vreeland (2008). Vreeland points out that the raw Polity data (Jaggers & Gurr, 1995) include explicit measures of political instability. 19 If these components are not removed, changes in a democracy score might reflect the security situation in a state rather than any institutional improvements or democratic decay. I record the post-civil war democracy level of each state, the year before the civil war and the year after the civil war ends, as well as from two to ten years after the end of the conflict. This measure is presented in Figure 1, and ranges from –6 to 7.
Potential confounding relationships
Despite previous ambiguous evidence on the correlates of post-conflict democratization, I utilize several research design strategies to measure and model potentially confounding relationships. Most simply, I also measure additional covariates that may impact postwar democratization but are unlikely to be themselves a function of rival aid during the civil war. 20 Following the research of Fortna & Huang (2012), I include two measures of economic development. First, I include the log of the prewar real gross domestic product of a state. I use the last observation before the civil war began. This is coded from Gleditsch (2002). I also include a measure of electricity consumption taken from Doyle & Sambanis (2000) to further differentiate countries that have a developed infrastructure that may support democratization as compared to less developed state contexts.
The number of factions involved in the civil war is also likely to affect the chances for post-conflict democratization, as argued by Wantchekon (2004). 21 While the relative size and strength of the factions might be an attractive covariate, it suffers from two problems. First, specific measurements of the military strength of rebel groups are not available globally. 22 Second, and most importantly, the causal mechanism we are interested in exploring, rebel aid during a civil war, directly influences the military ability of any faction that receives that aid. Including strength as either a matching (see below) or a control variable in a regression-based research design would risk introducing post-treatment bias. The number of factions in the conflict should be less susceptible to these problems since a group must be formed in order to receive aid. 23
Finally, I retained information on the role that United Nations intervention played in the civil war. Work by Fortna (2008) and Doyle & Sambanis (2000) emphasizes the positive but uncertain role that international organizations can play in post-conflict contexts. There is the strong suggestion of an effect in this work, although evidence has been scant on democratization specifically. 24
Estimating the quantity of interest
The key empirical question this article investigates is whether post-civil war democracy levels are systematically lower in countries where the executive is associated with a civil war group that received aid from an external state rival. There are several strategies I could use to estimate this impact, including regression, coarsened exact matching (Iacus, King & Porro, 2012), difference-in-difference estimation, and difference-and-difference matching (Heckman, Ichimura & Todd, 1997; Heckman et al.,1998). Due to space constraints, I review the uses of these strategies for making inference on the postwar effect of rival aid on democratization in the online appendix. 25
I focus initially on difference-in-difference matching that allows for some forms of both time-varying but observed heterogeneity and unobserved but time-constant unit heterogeneity, as well as common trends in democratization levels. This strategy uses the propensity score (see the appendix) to create a locally linear weighted average of untreated control units that each have pre-post values for democratization (Heckman et al., 1998). Then, subtracting the change in pre-post scores for the treated units and the change in the weighted pre-post scores for the untreated units provides the difference-in-difference matching estimate. The distribution of these estimates is calculated using a parametric bootstrap (Abadie & Imbens, 2006; Heckman et al., 1998). The balance across covariates is analyzed in the appendix, as is the assumption of common trends, which can be checked empirically using pre-civil war data.
Below, I present these difference-in-difference matching estimates and then compare them to parameters calculated from multivariate regression, regression-based difference-in-difference, and coarsened exact matching methods. While the difference-in-difference matching estimates may have the lowest bias relative to the other estimators here, they are also likely to have relatively high variance. If the stronger assumptions underlying regression, regression-based difference-in-difference or matching are met, these estimators may provide more precise estimates of the parameters of interest. Additionally, in investigating post-conflict democracy levels, I allow for distinct variances for the predicted democracy level at each time period and I do not assume a specific trajectory for the expected values. 26
Results
To explore the average effect of rival aid on postwar democratization trajectories, the difference-in-difference matching estimates and bootstrapped samples are presented in Figure 2. 27 The distribution of estimated parameters is illustrated with darker red indicating regions of higher density, meaning that the bootstrapped treatment effect is more likely to fall into this area. Light yellow and white areas represent areas of lower density, and thus regions where the treatment effect is less likely to have been estimated. A red horizontal line marks the point estimate, and the orange box fences the 90% bootstrap confidence interval. 28 The regions of highest density from years 2 to 10 are all below zero, indicating lower democracy scores for rival-aid post-conflict states, as predicted. For example, the xPolity score seven years after the civil war is estimated to be almost 3.5 points lower (90% CI: –6.4, 0.9) when the executive accepted aid from an external enemy. 29
This magnitude of change in xPolity score is meaningful when put in context. A change of almost 4 on the xPolity scale is nearly the difference between Zimbabwe in 1985 and Ireland in 1998 or between Rwanda and Malawi in the early 21st century. Given that the mean Heatmap of distribution of non-parametrically bootstrapped treatment effects from the difference-in-difference matching estimator. The thick line in the center of the distribution represents the mean; the box is the 90% confidence interval. Red/darker regions are higher density; yellow/light regions are lower density.
There are several additional pieces of evidence that lend credence to these results. First, as shown in the appendix, the local-linear weights improve the balance across the treated and untreated observations. While there are significant differences in the data on measures across GDP, development, the number of factions, and UN intervention for the treated and untreated groups before matching, after the weighting, these differences were not significant at conventional levels. Second, an analysis of the pre-civil war trends found no evidence that cases where a rival eventually provided aid had artificially high values of democracy pre-conflict. This mitigates the chance that the above estimates are biased downwards (see appendix).
Third, I have analyzed the data using several additional methods to probe the robustness of the findings. Specifically, I ran additional sets of analyses to estimate the difference in democracy level each year for states with rival aid versus those without. These results serve as a comparison to the difference-in-difference matched results. The first panel of Figure 3 presents an OLS regression analysis where democracy is assumed to be a linear function of the covariates. The same control variables were used here as in the matching equation. Second, I estimated a similar regression model but added controls from the xPolity score at the end of the war (in addition to the pre-war xPolity score that was previous included), the cost of the conflict in terms of battle deaths, and whether the war was an identity conflict or not. Third, I estimated the sample average treatment effect on the treated by using coarsened exact matching (CEM). Specifically, in the first stage: N matches are found for each treated observation using coarsened versions of GDP, development, number of factions, and appropriate weights calculated. 32 Then the matched data are analyzed and the original uncoarsened covariates are included as additional controls. These results are shown in the third panel. Finally, I estimated the regression-based un-matched differences-in-differences model. Thus, in the third panel, all observations are used, not just those that are matched. 33
Across the specifications, there are slight variations in the patterns, but the inferences are similar. Between four and five years after a civil war and continuing through ten years after the civil war ends, states with an executive Heatmap of distribution of estimated treatment effects from (a) regression, (b) regression with additional controls, (c) coarsened exact matching, and (d) difference-in-difference estimates. The thick line in the center of the distribution represents the mean; the box is the 90% confidence interval. Red/darker regions are higher density; yellow/light regions are lower density.
Additionally, I investigated several potential avenues where the relationship between executive rival aid and democratization would be spurious. It could be the case that any rival aid during a civil war, regardless of who receives it, decreases democracy scores. This would be the case if playing a spoiler was easily done by non-executives. I test this by including a variable measuring whether the losing side in the civil war received aid from a rival into the extended regression (with additional controls) previously presented in Figure 3. This does not alter the results. The effect for the sitting executive having received rival aid is negative and statistically significant, both from zero and as compared to the positive but insignificant coefficient on non-sitting executives having received rival aid during the war. 34

Top plot: the dots represent democracy levels at the end of the war and ten years after a civil war. The lines illustrate the linear trajectories. The line thickness and darkness represent changes of greater magnitude in absolute value. The dotted lines represent declines. Black represents countries with rival aid, white those without.
This means we are now in a position to recast the original puzzle of post-conflict democratization in Figure 1. Figure 4 splits the countries where the executive received rival aid during a civil war (black) and those that did not (white). 35 The top plot revisits the linear trajectories from the end of the war to ten years later, as the y-axis measures the level of democratization on the xPolity scale. Here we see that none of the states that ended up in the top right-hand corner, and thus were stronger democracies, were tainted by executives with rival aid. The only state that increased its democracy score after receiving rival aid by more than one point was Algeria, which moved from –3 to 2. However, in this case, the relationship between the Algerian government and its interstate rival Morocco remained tense even as they temporarily coordinated to deal with the rebel GIA organization. The limited cooperation occurred because both rivals viewed the GIA as a threat for a limited time. Further, there was significantly more conflict between the two states than there was cooperation. 36 On the other side, many of the large decreases in democracy level were states that received rival aid, as can be seen from the black dotted lines in Figure 4.
The differing trajectories of post-conflict democratization are more dramatically summarized in the lower plot in Figure 4. The black and white boxplots respectively illustrate the distribution of changes in democracy levels relative to the end of the war. In year 2 there are few changes, most are zero, but the only changes that occur for rival aid states are negative. By six years after the war, most non-rival aid cases are still around zero, while the distribution of rival democracy scores has moved downward. The means (denoted by x’s) further illustrate the pattern. Therefore, through the exploration of rival aid to executives, we have have gone some way towards explaining the puzzle of post-conflict democratization. Rival aid during the conflict forecasts a significant drag on post-conflict democratization, as predicted.
Conclusion
Taken together, these results both support and challenge existing notions concerning how actions taken during civil war alter the space for democratization in the postwar context. I find that one specific but stark civil war choice – accepting support from an external rival – reduces the scale of post-conflict democratization. Further, there is significant evidence that this drag on democracy can last for a decade, across different modeling assumptions and specifications.
This research supports the increasing relevance of work on transnational conflict processes. Gleditsch (2007) and Salehyan (2009) have established that relations between non-state actors and state actors across borders have significant relevance to the duration of a civil war, as well as its outcome. We can now add that transnational relations involving a rebel group can have echoing effects, possibly for a decade after the conflict ends.
In the future, it would be useful to merge this research with information on post-conflict elections as well as more specific measures of overt repression. It should be the case that these same leaders who are avoiding democratization should also be likely to attempt to undercut the legitimacy of elections, which they are likely to lose, and should be forced to mobilize the public through fear rather than institutional credibility.
These findings on the dark side of transnational relations also challenge the notion that the United Nations or other international organizations would be best to let states fight and exhaust themselves (Luttwak, 1999). While the exhaustion argument might have some appeal to those skeptical of intervention, it neglects the dual facts that civil wars do not occur in a vacuum solely penetrated by intermittent UN meddling, and that exhaustion is only one possible outcome. Cases where external rivals provided support within a civil war were split nearly evenly between conflicts where the UN was active (13 out of 34 cases, 38%) and where the UN was inactive or merely observing (48 out of 102 cases, 47%). However, the proportion of civil war groups that emerge from the civil war with control of the executive after receiving rival aid is zero when the UN plays more than an observer role (0 out of 34 cases). Conversely, almost 20% of conflicts without a significant UN role (more than observer status) led to executives in charge that had been associated with rival aid (20 out of 102 cases) and thus are less likely to democratize. The creation of a vacuum that is filled by transnational rival–rebel ties and long-term democratization deficits, as in Chad, is an example of this.
Footnotes
Replication data
The dataset and codes for the empirical analyses can be found at http://www.prio.no/jpr/datasets and at
. All analyses were conducted using R.
Acknowledgement
The author would like to thank the Journal of Peace Research editorial team and anonymous reviewers for their efforts in improving the article. Ben Appel, Kristian Gleditsch, Nils Metternich, Jakana Thomas, Henrik Urdal, Nils Weidmann, and Julian Wucherpfennig also deserve special thanks for their helpful suggestions. Any errors in the paper are the sole responsibility of the author.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
