CIVIL WAR IN SELF-DETERMINATION DISPUTES

We begin by presenting a brief theoretical discussion of the process leading to civil war in SD disputes. Civil wars are violent conflicts between a state and one or more organized non-state actors that reach some threshold of violence. They are typically either fought over control of the central government or some piece of territory in the state. In this article, we focus on SD disputes, so we only examine territorial incompatibilities.

SD disputes are the most common cause of civil war, but they are not the only cause. Indeed, several recent prominent civil wars, such as the wars in Syria, Afghanistan and Iraq, involve disputes over government. We focus on SD disputes because they provide a clear set of intrastate conflicts with the potential to become violent, but with variation among and within them over whether and when violence breaks out. As such, we can examine the potential for UNSC action to affect the likelihood of the onset of civil war. We expect, however, that this would also apply to disputes over government, as UNSC action would also have the potential to affect the decision making of states and dissidents in center-seeking civil wars. We return to a discussion of the potential differences between territorial and governmental civil wars in the conclusion.

Why do some SD disputes become civil wars and others do not? In general, disputes become civil wars when dissidents decide to use violence against state forces to pursue their aims. This focus on dissidents does not mean that states are rarely culpable; however, when states use violent repression, this is treated as a different phenomenon (such as one-sided violence or, at the extreme, genocide) unless dissidents respond with violence.Footnote ⁶

There are three main factors that affect whether dissidents choose to use violence against the state to pursue their aims. First, dissidents must perceive some potential benefit of doing so. The clearest way in which dissidents may benefit from violent rebellion is by winning the conflict. Through victory, rebels have the chance to impose their ideal outcome, which in a civil war stemming from an SD dispute could be the full secession of the territory in dispute.Footnote ⁷ Outright rebel victory in territorial conflicts, however, is extremely rare, and it is unlikely that many dissidents believe they can win the conflict. Yet even if they cannot win conflicts by defeating the government on the battlefield, dissidents can benefit from targeting violence against state forces if states grant concessions to them in return for an end to the conflict. In SD disputes, this often involves the state granting the SD group some degree of autonomy or control in the territory in which it is based. Negotiated outcomes are much more common than military victory by rebels, and are part of the process by which the rebels’ concerns and grievances are addressed.

Rebel groups can also perceive a benefit from fighting if it provides them with the ability to extract resources – natural resources and/or resources from local communities and economic actors – in areas they control.Footnote ⁸ These resources could include recruiting more support against the state. More broadly, the ‘outbidding’ logic in the terrorism literature suggests dissidents may have incentives to escalate to violent activity to gain politically within the constituent group by demonstrating their commitment to the cause.Footnote ⁹

A second variable that affects dissidents’ decisions to violently engage the state is the anticipated costs of doing so. Violent conflict is costly, as dissidents face the potential for violent repression and must exhaust resources in armed struggle.Footnote ¹⁰ Additionally, much of the fighting in SD disputes generally takes place in the territory under dispute, and destruction of this territory is costly for dissidents. Pertinent to costs, state strength is a key determinant of civil war onset.Footnote ¹¹ States vary dramatically in their ability to impose costs on dissidents. In some cases, the state is so weak that rebels can operate largely outside of its coercive reach – particularly in peripheral areas. In others, states are so strong that they can effectively repress violence throughout their territory by imposing large costs on dissidents.

The third factor that affects dissident decisions about whether to violently rebel is whether they and the government are able to reach compromises to avoid civil war when they recognize compromise to be mutually preferable to war. Reaching settlements can be difficult because of potential barriers to bargaining.Footnote ¹² Foremost, states and dissidents may be unable to reach compromise settlements that avert civil war if they do not agree on the likely course of that war. If states underestimate (or if dissidents overestimate) how costly conflicts will be for states, then the state may offer insufficient concessions, and civil wars may emerge as a consequence of the bargaining failure.

States and dissidents may also be unable to settle disputes if one or both sides cannot commit to abide by the terms of an agreement reached. If, for example, a state promises an SD group greater control over its own affairs, but the group does not believe that the state will follow through on that promise – especially if the group is vulnerable to prosecution or coercion and when high levels of mistrust persist – then dissidents within the group may choose to rebel. These commitment problems can lead to civil war even if states and dissidents have similar perceptions about the likely costs and outcome of the conflict. A related issue is that the parties may perceive an issue to be indivisible. In some cases – such as Kashmir or Jerusalem – the territory has strong symbolic importance for each side and so the actors may perceive the issue in all-or-nothing terms.Footnote ¹³

UNSC RESOLUTIONS AND CIVIL WARS OVER SELF-DETERMINATION

International actions, including those by the UNSC, can address these underlying causes of civil war and impact the likelihood of civil war. The UNSC is quite active in SD conflicts, and has passed a significant number of resolutions related to these disputes. The focus here is on conflict prevention. In general, UNSC preventive actions fall into four broad categories. First, the UN engages in diplomatic engagement – including good offices, mediation, fact-finding, civilian monitoring missions and the formation of special tribunals. For example, UNSC Resolution 367 of March 1975 called on the UN secretary-general to undertake a mediation role in the dispute over the status of Turkish Cypriots. Secondly, the UNSC authorizes the deployment of force, which can either include UN peacekeeping missions – such as the UN Observer Mission in Georgia established by Resolution 858 in August 1993 to supervise a ceasefire between the Georgian government and Abkhaz separatists – or authorizations of non-UN multinational forces – such as the NATO-led International Security Assistance Force in Afghanistan, which was authorized by Resolution 1386 in December 2001, and which, inter alia, affected the Uzbek and Tajik SD movements.Footnote ¹⁴ Thirdly, the UN levies sanctions on countries, and UNSC resolutions can impose, reauthorize or expand sanctions on the country involved. An example of this type of UNSC activity is Resolution 713, which established an embargo in September 1991 on military materiel being exported to the former Yugoslavia, where a number of SD disputes had escalated or were at risk of escalating to violence. Fourthly, and more passively, UNSC resolutions often contain condemnations, in which the Security Council deplores hostilities, non-compliance with prior UN directives or international agreements, or human rights violations. This last category includes Resolution 1076, issued in October 1996, which condemned the civil war in Afghanistan – a conflict that incorporated the Tajik and Uzbek SD struggles.

These types of actions related to UNSC resolutions have the potential to prevent SD disputes from escalating to full-blown civil war via different mechanisms. Broadly, UNSC resolutions – either through the resolution itself or through the actions that a resolution authorizes – are able to do this by increasing the costs of conflict for the combatants, decreasing the benefits of fighting (the opportunity costs of peace) or otherwise alleviating the barriers to bargaining.

DATA AND METHODOLOGY

To test whether UNSC resolutions decrease the likelihood of civil war, we use data from all years of an SD group’s existence from 1960 to 2005. The population of SD movements comes from the Center for International Development and Conflict Management (CIDCM) Peace and Conflict Report.Footnote ³⁸ This report identifies 145 SD movements residing in seventy-seven countries that demand more SD – that is, increased local control and autonomy for an ethnic group in a given territory.Footnote ³⁹ This list encompasses both groups that have participated in civil war, such as the Kurds in Turkey and Iraq, and those, such as the East Caprivians in Namibia, that have not. The data reflect the resolution or emergence of SD disputes at different times between 1960 and 2005. For example, the Afrikaners (in South Africa) dispute only emerges in 1994 with the end of apartheid, whereas the Kosovar Albanian dispute exits the data after 1999 with de facto Kosovar independence.Footnote ⁴⁰ We include a list of all SD groups in the Appendix, as well as an indication of whether the group experienced a civil war onset. Of the SD groups examined, 44 per cent experienced a civil war between 1960 and 2005. Table 1 shows the number and percentage of SD groups from each region as well as the percentage of groups in the region that experienced a civil war.

Table 1

Regional Distribution of SD Movements and Civil Wars

Our unit of analysis is the state-SD group dyad year. The dependent variable is an indicator of whether a new armed conflict, as defined by the Uppsala Conflict Data Project/Peace Research Institute of Oslo Armed Conflict Dataset,Footnote ⁴¹ began in that year. Armed conflict onsets are years when at least twenty-five battle-related fatalities occurred and in which there are at least two years of peace prior to the onset of hostilities. We use the twenty-five battle-deaths threshold instead of a higher threshold for two reasons. First, the mechanisms that we propose for the onset of intrastate armed conflict apply to both high and low levels of violence. To code the episodes of armed conflict with twenty-five to 1,000 battle deaths as periods of relative peace would miss important variations in the outcome of interest that we would like our model to explain. Secondly, we are interested in whether UNSC resolutions are able to proactively reduce the propensity for conflict, which would mean attenuating violence before a threshold like twenty-five battle-related fatalities. Efficacy at reducing the escalation to a much higher threshold without efficacy at lower levels of violence would imply a weaker – though still important – form of prevention that would be better characterized as reactive. In supplemental robustness checks, we find that the key findings at the lower threshold hold at the 1,000 battle-deaths threshold, suggesting that the preventive effects we find are not only applicable to the conflicts that were unlikely to escalate to massive violence.

We use Cunningham’s study of SD group fractionalization and the onset of armed conflict to tie the SD movements to specific armed conflicts.Footnote ⁴² Our analyses drop the years in which there is an ongoing armed conflict, as well as the two years immediately following the end of an armed conflict. Some of the onsets are new armed conflicts, while others are recurrences. We conduct different analyses with different samples to assess whether UNSC resolutions help prevent the onset of new armed conflicts involving previously peaceful SD movements, the recurrence of armed conflict involving SD movements or both. One sample thus includes all groups at risk for conflict, and another sample includes just those that have yet to experience a civil war.

Using discrete binary time-series cross-sectional data, we run logit regressions that are essentially event-history models.Footnote ⁴³ To account for duration dependence, we follow Carter and Signorino and include a cubic polynomial of the time at risk.Footnote ⁴⁴ Since our analysis begins in 1960, but many of the groups existed prior to that year, we would run into the problem of left-censoring if we started the risk time for all groups at zero in 1960.Footnote ⁴⁵ So, for the thirty-four groups that already existed in 1960, we chose as their origin point the later of the following: (1) the year after any post-1946 armed-conflict episodes ended; (2) the entrance into the international system of the respective country or (3) 1946.Footnote ⁴⁶ We report standard errors that are robust to clustering on the SD movement.

Our independent variables measure the content of recent UNSC resolutions related to the SD movement. Beardsley has coded information on the content of all the resolutions during the time period of our study, including which countries were the concern of the resolutions and the actions mandated in the operative paragraphs – condemnations, diplomacy, force or sanctions.Footnote ⁴⁷ After matching the resolutions to all the relevant countries with SD movements, we went through each resolution and determined whether the resolution directly pertained to the SD movement in question, indirectly pertained or was not relevant.Footnote ⁴⁸

In the analyses below, we examine whether different types of UNSC resolutions have different effects. We aggregate the various types of activities in the resolutions into the following categories: diplomacy, new force authorizations and mandate expansions, sanctions, force re-authorizations, condemnations and ‘all else’. The categories are not mutually exclusive, except that the ‘all else’ category is true only if there is no activity of the other listed types.Footnote ⁴⁹ ‘Diplomacy’ includes UNSC instructions for the secretary-general to offer his good offices, mediate or investigate an issue; mandates that empower fact-finding or monitoring entities; calls for other actors to mediate; the formation of special courts; and mandates that call for UN entities to carry out peacebuilding activities – focused on strengthening political and economic institutions – including demobilization, disarmament and reintegration programs. Although some of these actions are distinct from each other, they share an emphasis on non-coercive UN involvement. Moreover, Boutros-Ghali’s Agenda for Peace similarly combines these activities as part of ‘preventive diplomacy’. ‘Force’ mandates include those pertaining to missions overseen by the UN Department of Peacekeeping Operations, as well as those carried out by multinational forces with explicit authorization from the UNSC. Mandates of ‘new force’ are separated from ‘force reauthorizations’ because, unlike other types of UN action, force authorizations carry an explicit expiration date that can only be extended with another resolution. Simple extensions of mandates to a later date are coded as their own category, and are generally considered to be involvement that is not nearly as active as resolutions that authorize or endorse new approaches. We include mandate expansions and all authorizations to use all necessary means as new force. ‘Sanctions’ are explicit demands and authorizations for the international community to restrict the trade, financial privileges, or movement of specific governments or leaders. Resolutions are coded for sanctions when they authorize new sanctions as well as when they call for the continuation or expansion of existing sanctions. Finally, ‘condemnation’ covers explicit statements in the operative paragraphs that deplore such actions as hostilities, human rights abuses and the failure to comply with existing agreements. We also include in this category explicit threats – but not the actual implementation – of sanctions or other punishments.

We then created two separate indicators for each type of action: one for the resolutions that directly pertained to the SD movement in question, and one for the resolutions that indirectly pertained. In most cases, resolutions were determined to directly pertain if the SD movement was explicitly mentioned. For example, Resolution 541 (1983) reads ‘The Security Council […] Deplores the declaration of the Turkish Cypriot authorities of the purported secession of part of the Republic of Cyprus’ (emphasis added). The direct relevance to the Turkish Cypriot SD dispute is clear. However, many resolutions were determined to have direct relevance based on the context in which they were issued. The overarching decision rule for direct pertinence was whether the issuers of the resolution (the UNSC) and the targets (the government in question and the SD movement) would have understood the resolution to primarily pertain to them, even if they were not explicitly mentioned. Where that criterion was satisfied, the resolution was coded as directly pertinent. For example, Resolution 383 (1975), which renewed the peacekeeping mandate of the UN mission in Cyprus and called for renewed peace talks between the ‘two communities’, was coded as pertaining to direct force, even though the ‘Turkish Cypriots’ are not explicitly mentioned. Given the context in which the resolution was passed, ‘two communities’ could only be understood as referring to Greek and Turkish Cypriots. Resolutions pertaining generally to Israeli incursions into southern Lebanon in the 1970s (for example, Resolution 425 in 1978) were understood to directly pertain to Palestinians living in Lebanon – even though they did not name ‘Palestinians’ – as the Israeli incursions were primarily directed against Palestinian militant groups.

Resolutions that did not meet the criteria for direct relevance – meaning that they did not explicitly mention or primarily pertain to the SD movement, but addressed a broader statewide or regional conflict to which the SD group was a plausible participant – were coded as indirectly relevant. For example, resolutions during the Congo crisis in 1960 that addressed the secessionist conflict in Katanga Province (for example, Resolutions 143 and 145) were coded as indirectly pertaining to the Lunda and Yeke SD movement, because this group was heavily represented in Katanga Province and was therefore a plausible participant in the Katanga conflict. The resolutions were not coded as being directly relevant to the Lunda and Yeke, because they were not explicitly mentioned in the resolutions, nor was the conflict specifically a manifestation of the Lunda and Yeke SD movement. The Katanga resolutions, meanwhile, were coded as being neither directly nor indirectly relevant to another Congolese SD group, the Bakongo. The Bakongo were concentrated far away from Katanga (in a country roughly the size of Western Europe) and were therefore not a plausible actor in the Katanga conflict.

In another example, broad condemnations of apartheid policies in South Africa (for example, Resolution 391, 1976) were coded as indirectly relevant to both the Zulu and Khoisan SD movements because the conflict over the apartheid system included the formation of Bantustan ‘homelands’ for these groups, making them potential participants in the anti-apartheid conflict. These resolutions were not directly relevant to these SD movements because the Khoisan/Zulu SD movements were not well integrated with the anti-apartheid social movements, and were thus not directly implied as pertinent to the UNSC resolutions against apartheid.

Also, resolutions that addressed an international conflict with inherent relevance to SD movements were coded as indirectly related. For example, the establishment of a peacekeeping mission in the Golan Heights (Resolution 350, 1973) was coded as indirectly related to the Palestinian SD movement in Israel, given the Palestinians’ role in the broader Arab-Israeli conflict as potential combatants and as a galvanizing cause. In sum, if a resolution did not explicitly address an SD movement and could not be interpreted as being primarily addressed to that SD movement, but did address a broader state- or region-wide conflict to which the SD group was a plausible participant – regardless of their actual involvement – the resolution was coded as indirectly relevant.

Resolutions that did not fulfill the criteria for either direct or indirect relevance were classified as not relevant – even if they pertained to the country in which the SD movement resides. Resolution 164 (1961) provides a clear example. This resolution addressed French colonial policy in Tunisia, and accordingly was coded as having no relevance to the Basque and Corsican SD movements in metropolitan France.

Of the 142 SD movements at risk of armed conflict in our sample, nineteen had some directly pertinent UNSC resolution condemning the actors – nine had sanctions, twelve had new force authorizations, nine had re-authorizations, eleven had instructions for diplomacy and twenty-one had resolutions in the ‘all else’ category. Twenty-five had some indirectly pertinent UNSC resolution condemning the actors – nineteen had sanctions, seventeen had new force authorizations, fifteen had re-authorizations, twenty-four had instructions for diplomacy and twenty-nine had resolutions in the ‘all else’ category.Footnote ⁵⁰

To measure the effect of recent UNSC resolutions, we use an exponential decay function. We posit that the effect of a resolution is likely to last much longer than a single year, but that the weight of its influence should decrease over time. Moreover, when multiple resolutions of the same type have been issued in the recent past, they should have a cumulative effect. Using a decay function allows us to capture these dynamics. We use a four-year half-life, in which a resolution has a weight of 1 in the year in which it was adopted, and the weight decreases each year such that it has a weight of 0.5 after four years, 0.25 after eight years, etc.Footnote ⁵¹ As additional resolutions are adopted, they add a weight of 1 to the decaying measure, and a new decay function is applied to the weight of that resolution at its date of adoption. Multiple resolutions with the same actions in the same year are treated as having an initial weight equal to the number of resolutions. To avoid simultaneity bias, we lag all the independent variables by one year.

We perform separate regressions for each type of UNSC action while controlling for the combined weights of all the other types of action. The estimated coefficients on the actions variables should thus be interpreted as the effects of those actions while holding constant the general level of other UNSC involvement. In addition to controlling for the time at risk, we control for a number of variables that correlate with the ex ante fragility of peace. If UNSC involvement is fairly reactive and tends to go to the cases at greatest risk of major violence,Footnote ⁵² then the analysis will tend to understate the UN’s ability to prevent armed conflict.

Many of our control variables follow from Cunningham.Footnote ⁵³ We control for the number of SD movement factions (logged), which Cunningham shows to robustly increase the chance of civil war. Cunningham also finds that previous concessions play a role in the outbreak of violent conflict, and we include this measure to control for the history of conflict management between movements and governments. We include a binary indicator of democracy in the models, since democratic states have more legitimate political avenues by which the SD group’s grievances can be heard, and therefore are less likely to experience civil war.Footnote ⁵⁴ The presence of ethnic kin groups in neighboring states has been shown to increase the propensity for armed conflict, and so we include a binary indicator of such groups.Footnote ⁵⁵ Gross domestic product (GDP) per capita has a robust correlation with intrastate conflict,Footnote ⁵⁶ as does the number of active SD groups.Footnote ⁵⁷ Finally, we control for the average amount of economic discrimination in the country from the Minorities at Risk data, since Cederman, Gleditsch and Buhaug find that discrimination in ethnic conflicts shapes the propensity for political violence.Footnote ⁵⁸

In addition to these core control variables, we conducted additional robustness checks by adding additional regional temporal control variables in order to account for other possible confounds. Specifically, we included dummy variables for sub-Saharan Africa, Asia and Eastern Europe,Footnote ⁵⁹ as well as a dummy variable for the Cold War. The key results are robust to these extensions and are reported in the Appendix (Table A3). We also include in robustness models a measure of whether peacekeeping operations, from the International Military Interventions data, are deployed to the respective countries (Table A4). We do not include this as a core control variable because it is an intermediate variable between the force authorization measures and the outcomes of interest. That being said, it is telling that the findings are similar with and without this variable, which suggests that the effect of the force mandates does not solely operate through the deployment of troops on the ground, and that there is likely to be a broader signaling effect of strong UNSC action. Finally, we also considered the possibility that the estimated size of the group as a percentage of the state’s population and the presence of hydrocarbon reserves might shape both regime-rebel incentives and the UNSC response. The results are robust to the inclusion of these variables, and they have little independent effect on the outcome (Table A5).

Table 2 presents the regression results for our models of the more active types of UN activity. Due to space constraints, the results for the more passive types of UN involvement – condemnations and simple reauthorizations of force – are shown in the Appendix and indicate that such actions do not have a statistically significant relationship with the onset of conflict (Table A2). Our theoretical discussion highlighted the potential distinction between UNSC resolutions that directly and indirectly pertain to SD movements. We discuss the evidence for each of these mechanisms in sequence below.Footnote ⁶⁰

Table 2

Logit Models of Armed Conflict Onset

Note: robust standard errors in parentheses. Squared and cubed risk time not shown (see Appendix Table A6). ***p<0.01, **p<0.05, *p<0.10, two-tailed test.

FURTHER ANALYSES

While the observed deterrent effects of indirectly relevant authorizations of new force and sanctions are not likely to be explained by selection effects, the findings related to direct diplomacy and direct sanctions warrant additional scrutiny. If the process by which the UN chooses to directly act is related to the likelihood of civil war, then our results may be biased. Consider two possibilities: if the UN chooses to become involved in the easiest disputes – those in which the likelihood of civil war is the lowest – then the negative correlation between direct diplomacy and civil war may just indicate that these disputes were unlikely to escalate to civil war anyway, not that diplomacy has any effect. Yet if the UN chooses to become involved in the disputes with the highest likelihood of civil war, than a positive correlation between direct sanctions and civil war may simply indicate that sanctions are used in the most severe disputes, not that they increase the likelihood of civil war on their own.Footnote ⁶⁵ Empirically, in examining how our control variables correlate with the types of intervention, we do not observe that diplomacy with direct relevance to an SD movement tends to be applied to the easier cases, or that sanctions with direct relevance tend to go to the more difficult cases.Footnote ⁶⁶

To further analyze the direct effects of diplomatic involvement and sanctions, we also conducted seemingly unrelated regression models to simultaneously estimate an equation of these types of direct UN involvement and the outcome equation while controlling for the correlation in errors between the equations.Footnote ⁶⁷ In doing so, we control for some of the processes left outside the model that potentially shape both how active the UN is in addressing an SD dispute and the ultimate potential for violence to escalate. We used dummies for the Cold War, Western states, P-5 status and P-5 allies as variables in the UN-involvement equation that did not appear in the outcome equations.Footnote ⁶⁸ These dummy variables are likely to affect the political feasibility of UNSC actionFootnote ⁶⁹ without strong theoretical connections with the escalation of violence related to an SD movement.

In these regressions (Table A7), the correlations in the errors across the equations are not statistically significant, and the estimated relationships between conflict onset and both direct diplomatic activity and direct sanctions remain statistically significant. We therefore rely on the logit models in Table 2 for our analyses. The robustness of the finding that direct sanctions are associated with a higher likelihood of civil war remains puzzling, given that a moral hazard explanation is difficult to justify in light of the observation that indirect sanctions do not have a similar relationship with civil war onset. Further exploration reveals that four of the five cases of civil war onset when the UNSC has authorized sanctions in the recent past come from the Balkans. While the failure of the sanctions regime in the former Yugoslavia is well documented,Footnote ⁷⁰ we hesitate to generalize too much from what is essentially a single case.