Electoral competition

The sine qua non of democracy is the holding of regular, competitive, national elections. When the people choose their leaders, incumbents have an incentive to keep foreign policy decisions closer in line with the majority of the electorate (Bueno de Mesquita et al. Reference Bueno de Mesquita1999). Indeed, research shows that mistakes in the international arena can have consequences not only for results at regularly occurring elections but also for popular support between elections (Bueno de Mesquita and Siverson Reference Bueno de Mesquita and Siverson1995; Gelpi Reference Gelpi2017: 1932–1936). Thus, popular discontent with foreign policy decisions may have a direct bearing not only on the political survival of a democratic government but also on its general political manoeuvrability (Tomz and Weeks Reference Tomz and Weeks2013: 850).

We argue that the shadow of competitive elections also has ramifications for how democratic leaders view interventions in intra-state conflicts abroad. Fighting wars is always difficult to justify. Numerous studies show that citizens are sensitive to costs, especially casualties, when forming opinion on whether to support or oppose engagement in armed conflicts (Gartner Reference Gartner2008; Gelpi et al. Reference Gelpi, Feaver and Reifler2009; Karol and Miguel Reference Karol and Miguel2007). Likewise, we know that support for military actions plummets if voters do not feel that a mission is necessary to uphold national or international security (Hayes Reference Hayes2009; Russett Reference Russett1994: 38–39). In this light, interventions in civil wars should be particularly difficult to justify because the survival of the nation is rarely at stake. Furthermore, as interventions in civil wars abroad often become costly and messy affairs, we expect public dissatisfaction to be a real concern for decision-makers within democratic third-party states.

We do not argue that electorates in democracies see no benefit in civil war interventions at all. For example, voters might support interventions that aim to promote democracy or stop mass atrocities. We simply argue that such concerns seldom outweigh the costs of intervention for a majority of the electorate. In other words, the majority is rarely ready to fight for these objectives if it means that the nation must incur substantial costs in the form of casualties or a reduction of spending on welfare (De Mesquita and Downs Reference De Mesquita and Downs2006: 631).

Autocrats do not have the same constraints. This is not to say that autocrats do not face audience costs (Lai and Slater Reference Lai and Slater2006; Weeks Reference Weeks2012), nor that the relevant audiences do not affect decisions concerning third-party interventions (Kinne Reference Kinne2005). However, as authoritarian rulers can manipulate elections (or cancel them altogether) and stifle public opposition through repression, the burden of vertical accountability weighs much more lightly on their shoulders. In contrast to democratic governments, they need not fear repercussions from a disgruntled public and therefore rarely devote the same concern to avoiding potentially costly and difficult-to-justify interventions in conflicts abroad.

Checks on the executive

Democratic leaders also face more horizontal constraints than their autocratic counterparts – this also applies when making decisions about intervention in civil wars. In many democracies, the parliament must first consent to the deployment of troops abroad or approve budgets for military operations (Maoz and Russett Reference Maoz and Russett1993: 625–626). Furthermore, once approved, spending will remain under the scrutiny of parliamentarians, non-partisan bureaucrats and audit agencies. For example, in the US, the president can initiate a civil war intervention on his own, but both houses of Congress must approve military budgets and often stipulate how money is spent. Similarly, in a number of European countries, including Austria, Denmark, Ireland and Spain, the government cannot send troops abroad without the prior passing of a regular law in parliament (Born et al. Reference Born, Fuior and Lazzarini2007). These formal institutional constraints are important in themselves. But when coupled with the vertical constraints and the expected public reluctance towards civil war interventions discussed above, they are likely to bind even tighter (Tomz and Weeks Reference Tomz and Weeks2013: 850). When opposition parties and candidates, and even members of the government parties, sense defiance among the general public, they are likely to make more active use of their veto powers to capitalize politically (Gelpi Reference Gelpi2017: 1936).

Autocrats, on the contrary, only need support from the most powerful people in their ruling coalitions – and most often without the hassle of formalized decision-making processes (Svolik Reference Svolik2012: 197). In addition, constitutional stipulations that constrain the executive in the foreign policy domain are frequently either ignored or done away with over time. The recent Russian intervention in the Syrian civil war provides a telling example. The surprise engagement in September 2015 was preceded neither by discussion of the issue in public media nor by debate in the Russian parliament. Moreover, once President Vladimir Putin announced the decision, parliamentarians willingly played along and applauded every aspect of the operation (Balmforth Reference Balmforth2015). Putin never had to justify or win the broad support for the decision to send Russian troops to far-away Syria to fight for the survival of the oppressive Assad regime. Without the constraint of horizontal accountability that democratic governments face, the freedom to engage in third-party interventions is much wider for authoritarian leaders.

Critical media

A third defining character of democracies is a free and pluralistic media. This particular institution has two consequences for decisions concerning third-party interventions that feed back into the first two mechanisms presented above. First, it makes it more likely that the financial, humanitarian and political downsides of an ongoing – or a potential future – intervention come to the awareness of the public and the opposition (Fearon Reference Fearon1997; Schultz Reference Schultz1998, Reference Schultz1999). Indeed, experiments show that citizens do update their attitudes towards a conflict if exposed to new, relevant information (Gartner and Gelpi Reference Gartner and Gelpi2016). Knowledge of negative consequences of an intervention may therefore spur or fuel popular discontent and it can inspire the opposition to politicize the issue and perhaps make use of their formal veto powers. Thus, a critical media can augment the effect of the first two mechanisms. Second, the existence of a free and critical media forces political leaders to consider carefully how they justify an intervention. Motivations such as protecting oil interests, supporting a corrupt government in order to maintain an important ally, or deliberately prolonging a civil war to weaken a foe will not be an easy sell in a democratic context.

In combination, these two concerns mean that democratic leaders face a higher risk of attracting undesirable, critical attention from political opponents and the public at large in connection with interventions abroad. The anticipation of incurring reputational and political costs as a result of negative media attention probably makes democratic leaders more hesitant to intervene in the first place. President Clinton's hasty decision in October 1993 to withdraw US forces from Somalia exemplifies the costs that democratic leaders fear to incur. Not only did the pictures of a US soldier being dragged through the streets of Mogadishu outrage the general public, they also led to heated recriminations in Congress over the wisdom of the intervention (Bolton Reference Bolton1994). Indeed, ever since, the Somali incident has figured prominently in the public debate as a reminder of the potential high costs associated with interventions (Brunk Reference Brunk2008), thus illustrating how an open media environment constrains democratic leaders’ decisions about interventions.

On the contrary, most autocracies do not have an open discussion on foreign policy issues nor an independent and pluralistic media (Stier Reference Stier2015). Usually, state-controlled media and the majority of the political establishment follow the official line of the regime (Stockmann and Gallagher Reference Stockmann and Gallagher2011). If opposition is voiced, it is normally done so by marginalized regime critics or by smaller media outlets with minimal impact on public opinion (for an example, see LaPorte Reference LaPorte2015: 349). Consequently, authoritarian leaders need not fear that details concerning the negative consequences of a civil war intervention come to the awareness of the public nor that their justification of the intervention will be seriously contested. Again, the Russian support to President Assad in the ongoing civil war in Syria provides a telling example. Not only does the Putin regime consistently manipulate the official death toll (Reuters 2017), it also refuses to publish figures on the costs of its operation (Bocharova et al. Reference Bocharova, Mukametshina and Nikolski2017). Nonetheless, the dominant media within Russia remain silent.

To recapitulate: we expect countries with democratic institutions to be less inclined than authoritarian countries to send support to intra-state armed conflicts abroad. The three mechanisms we have presented each work independently, but they also reinforce each other and thus strengthen the relationship further. On this basis, we form our first, primary hypothesis:

Types of third-party support

We should not expect democratic leaders to be only less likely to initiate civil war interventions in general. In the following, we argue that the three same mechanisms also constrain democratic leaders in terms of how they engage in interventions when they do decide to interfere. Specifically, we expect that democracies will be less likely to provide combat-intensive types of support, such as sending combat troops and weapons, as opposed to providing more indirect support types such as training, intelligence, logistical support and so on.

Sending combat troops to fight alongside one or more warring parties in an intra-state conflict is arguably the most visible and often most costly type of support a third-party state can provide. However, the visibility and the perceived costliness of such an involvement is highly dependent on the strength of the three institutional constraints emphasized in this article. As the Russian intervention in Syria shows, autocrats can deploy and maintain troops without public discussion of the wisdom and necessity of such an involvement and without the nagging criticism of opposition politicians. In contrast, democratic leaders must always be ready to face the anger of the electorate and the scrutiny of the opposition and the media when the country's boys are brought home in coffins (Gartner Reference Gartner2008; Gelpi et al. Reference Gelpi, Feaver and Reifler2009; Karol and Miguel Reference Karol and Miguel2007).

Provision of weapons is also a highly visible support type. Though democratic leaders do not have to worry about casualties on their own side when delivering weapons to warring parties, the provision of the equipment necessary for killing opponents and destroying infrastructure does not square well with democratic ideals of non-violent conflict resolution, just war and the protection of civilians (Christensen Reference Christensen2015). Again, this means that democratic leaders must invest heavily in justifying such partisan support and not least defending the actions and goals of those groups that receive the weapons (Schmitt Reference Schmitt2014). The protracted discussion of whether or not to send arms to the Ukrainian government in its fight against Russian-sponsored separatists in the east indicates that such dynamics of constraints clearly influence decision-makers in democracies (Schmitt and Kramer Reference Schmitt and Kramer2017). Again, due to the stifled political context and the much weaker vertical accountability, rulers in autocracies do not face the same needs for public justification, and often the provision of weapons can effectively be kept secret from the public.

Turning to more indirect support types such as access to territory or provision of logistics, intelligence, training and expertise, the expectations are less clear-cut. Indirect support types are less likely to be financially burdensome, less likely to result in casualties and are rarely subject to prior formal approval in parliament – all of which make indirect support less problematic for democracies. Still, even indirect support – if poorly justified – can attract the attention of the opposition or trigger public debate. On this basis, we still expect democratic governments to be different from their authoritarian counterparts, but less so than for the other two support types. Thus, we formulate two additional hypotheses:

Dependent variable

To identify third-party support from foreign states in intra-state armed conflicts we use the UCDP External Actor Dataset (UESD) (Högbladh et al. Reference Högbladh, Pettersson and Themnér2011),Footnote ⁵ which contains disaggregated information on 10 different types of third-party support to all warring parties listed in the UCDP/PRIO Armed Conflict Dataset. The UESD is thus more comprehensive than the data sets previously used in the literature; namely, Regan's intervention data (Regan Reference Regan2002) and the International Military Interventions Dataset (Pearson and Baumann Reference Pearson and Baumann1993; Pickering and Kisangani Reference Pickering and Kisangani2009).

External support is defined as ‘warring or non-warring support to a primary party that is given to assist it in an ongoing armed conflict’ (Högbladh et al. Reference Högbladh, Pettersson and Themnér2011: 5). In line with our theoretical focus, the data only include partisan interventions, where support is given with the intention to strengthen a party in the conflict. This also means that UN-sponsored peacekeeping operations and ‘action against foreign rebel groups that are merely carried out to protect one's sovereignty’ (Högbladh et al. Reference Högbladh, Pettersson and Themnér2011: 5) are left out. The UESD contains 145 intra-state armed conflicts in 89 countries, totalling 973 conflict country years. In almost three-quarters of these conflicts (108 out of 145), state-sponsored third-party support was provided to one or more of the warring parties.Footnote ⁶ When reorganized into dyads of conflict country and potential intervener country, third-party support is present in 2,356 (1.51%) dyad years out of the total 155,750 dyad years in the period 1975–2009.Footnote ⁷

To test hypotheses H2a and H2b, concerning the type of third-party support provided, we combine the UESD's 10 different support types into three overall categories that match our theoretical distinction between combat-intensive support and indirect combat support. In the former category, we distinguish between troop support, which includes all types of troops fighting alongside one of the warring parties, and weapons support, which includes donations, transfers and loans of weapons and ammunition (Högbladh et al. Reference Högbladh, Pettersson and Themnér2011: 6–8). In the latter category, we include funding/economic support such as military loans and grants, access to intelligence material, material/logistics such as vehicles and uniforms, access to territory, access to military/intelligence infrastructure and training/expertise (Högbladh et al. Reference Högbladh, Pettersson and Themnér2011: 6–8).Footnote ⁸ In years where a third-party state is involved in several types of support to one or more warring parties in the same country, we classify cases hierarchically.Footnote ⁹ Based on this coding, weapons support is the most common support type (963 dyad years), while troop support is the least common support type (617 dyad years). Cases of pure indirect support fall in between the two others (776 dyad years). In the online Appendix, we present results with alternative categorizations of support.

Table 1 lists the 10 most frequent sender countries and target countries of third-party support based on all 2,356 dyads of support in the data. The US is, not surprisingly, an exceptionally active supporter involved in almost three times as many dyads of support as the second-most active third-party state, Russia (the Soviet Union). More importantly, we note that eight out of 10 of the most frequent supporter countries are autocracies. For example, Cuba under Fidel Castro supported warring parties in several Latin American and African conflicts during the Cold War period (e.g. Nicaragua, El Salvador, Ethiopia, Angola, Mozambique, Zimbabwe and South Africa) with weapons, training and sometimes even combat troops (Angola and Ethiopia). Libya under Muammar Gaddafi was another very active third-party supporter, especially in North Africa. Finally, Iran and Syria have been heavily engaged in support throughout the Middle East. Among less frequent interveners we find several autocracies but also some democracies including the UK, Israel, Italy, Australia and India.

Table 1.

Top 10 Third-Party Supporters and Target Countries 1975–2009

Note: Observations are dyad years of support.

Turning to the most frequent targets of support, Iraq and Afghanistan top the list. The majority of dyad years of support for these countries relate to the conflicts following the US-led invasions in 2001 and 2003, in which a substantial number of countries, mostly democracies, provided third-party support.Footnote ¹⁰ Somewhat surprisingly, the US itself is also present on the list. The reason is that after the 9/11 terrorist attacks, the US government fought al-Qaeda on its own territory, and during this conflict Washington received ample support from its allies abroad. Again, by far most of these third-party supporters were democracies. It is debatable whether these three armed conflicts connected to the War on Terror, Afghanistan (2001–), Iraq (2003–) and the US (2001–), should be considered as intra-state armed conflicts, as they are clearly qualitatively different from the other 142 intra-state armed conflicts in the data set. The issue is particular pertinent for this study because these special conflicts are also characterized by a very large number of instances of third-party support. We choose to exclude all three conflicts from our main analyses but also run a number of robustness checks where they are included (see below). Beyond the most frequent targets of support, which are located in Africa and the Middle East, the data include target countries from all parts of the world such as Cambodia, Nicaragua, Philippines, Azerbaijan, Sri Lanka and Argentina.

Independent variable: political regime type

Following the convention in conflict studies, we identify regime type using the revised polity2 scale from the Polity IV data set (Marshall et al. Reference Marshall, Gurr and Jaggers2017).Footnote ¹¹ The six components, from which the polity2 scale is constructed, directly or indirectly reflect the regime-type characteristics that we emphasized in our theory.Footnote ¹² To test our hypotheses, we create a set of dummy variables that distinguish autocracies (polity2 score −10 to −6) from anocracies (polity2 score −5 to 5) and democracies (polity2 score 6 to 10). In our data, 60,196 dyad years (41.8%) have a democratic potential intervener, 50,185 dyad years (34.9%) have an autocratic potential intervener, and 33,477 dyad years (23.3%) have an anocratic potential intervener. We also present results using the continuous 21-point polity2 scale (rescaled to range from 1 to 21).

We are aware of the criticism raised against the polity2 scale (Cheibub et al. Reference Cheibub, Gandhi and Vreeland2010; Treier and Jackman Reference Treier and Jackman2008). To make sure that our results do not depend on the Polity data itself, the online Appendix presents robustness checks with an alternative aggregate measure of democracy from the Varieties of Democracy (V-Dem) data set (Coppedge et al. Reference Coppedge2016).

Control variables

We control for conflict country confounders, supporter country confounders and dyadic confounders. As for the former, we include three controls. The first is the conflict country's total annual oil production in metric tonnes (Ross and Mahdavi Reference Ross and Mahdavi2015) since access to natural resources, especially oil, is argued to attract third-parties in general and autocracies in particular (Bove et al. Reference Bove, Gleditsch and Sekeris2016; Koga Reference Koga2011). We also include a control for rebel group strength since democracies might be particularly prone to provide support to strong rebel groups (Koga Reference Koga2011: 1146–1147) and since rebel group strength might affect the probability of external support in general.Footnote ¹³ We use the NSA data set (Cunningham et al. Reference Cunningham, Gleditsch and Salehyan2013) to identify the strongest rebel group vis-à-vis the government on an ordinal scale ranging from 1 (much weaker than the government) to 5 (much stronger than the government).Footnote ¹⁴ Lastly, we control for the conflict country's regime type, polity (conflict country), using the full 21-point polity2 scale (Marshall et al. Reference Marshall, Gurr and Jaggers2017). The latter ensures that our results are not driven by democracies only intervening in democratic countries, which typically experience fewer intra-state conflicts, and autocracies only intervening in autocratic countries, which have more intra-state conflicts.

Turning to supporter country confounders, we include a control for major power status, since major powers are more likely to intervene (Corbetta and Dixon Reference Corbetta and Dixon2005) and since most major powers are democracies. Using the Correlates of War State System Data, we create a dummy variable indicating whether a country was considered a major power (1) or not (0) (Correlates of War Project 2016).

We control for five potential dyadic confounders. Using the ICOW Colonial History Data Set (Hensel Reference Hensel2014),Footnote ¹⁵ we create a dummy variable, colonial relations, indicating whether the conflict country was a former colony of the potential intervener country (1) or not (0). Intervention is more likely in a former colony because of political and cultural linkages (Koga Reference Koga2011; Lemke and Regan Reference Lemke, Regan and Diehl2004; Stojek and Chacha Reference Stojek and Chacha2015), and most countries with prior colonial possessions are today democracies. To capture alliance bonds, we use the Correlates of War Formal Alliances Dataset to create a dummy variable, alliance, indicating whether any type of alliance exists between the conflict country and the potential intervener country (1) or not (0) (Gibler Reference Gibler2009).Footnote ¹⁶

Since countries often shy away from involvement in other countries that are stronger than themselves, and since democracies often have stronger capabilities than autocracies, we also include a control for the capabilities ratio between the two countries (Bove et al. Reference Bove, Gleditsch and Sekeris2016; Corbetta and Dixon Reference Corbetta and Dixon2005; Findley and Teo Reference Findley and Teo2006; Koga Reference Koga2011). We divide the Composite Index of National Capability (CINC) score of the conflict country by the CINC score of the potential intervener country using the Correlates of War National Material Capabilities Dataset (Singer et al. Reference Singer, Bremer, Stuckey and Russett1972).Footnote ¹⁷ We also control for the distance between the borders of the conflict country and the potential intervener country in kilometres at the closest point using the CShapes data set (Weidmann et al. Reference Weidmann, Kuse and Gleditsch2010). Most studies of intervention include a control for distance since the cost of intervention rises the further away a conflict is (Bove et al. Reference Bove, Gleditsch and Sekeris2016; Koga Reference Koga2011; Stojek and Chacha Reference Stojek and Chacha2015). However, distance is particularly important in our analysis because civil wars tend to cluster in regions with more autocracies. Thus, failing to control for distance might lead us to overestimate the effect of democracy. Finally we include a control for the total sum of bilateral trade in millions of current US dollars from the COW bilateral trade data (Barbieri et al. Reference Barbieri, Keshk and Pollins2009),Footnote ¹⁸ since democracies often have higher levels of trade, and since trade has been argued to increase the probability of intervention (Stojek and Chacha Reference Stojek and Chacha2015).

Lastly, we introduce two time controls. First, we control for differences between the Cold War period and the post-Cold War period as third-party interventions could be expected to be more frequent during the geopolitically contested Cold War period (Regan Reference Regan1998: 767), which might coincide with changing incentives to intervene for democracies and autocracies. We construct a dummy variable, Cold War, which takes the value of 1 until 1989 and 0 from 1990 and onwards. Second, to control for omitted variable bias correlated with time, we include time since intervention, which counts the years since the last intervention in the same dyad.Footnote ¹⁹ For example, a country that has recently been engaged in an intervention might have more knowledge concerning local contexts and challenges, which could increase the likelihood of conducting another intervention. We lag the variables alliance, bilateral trade, polity (conflict country) and oil production one year, since they may be affected by the provision of third-party support. We also log-transform the continuous control variables capabilities ratio, bilateral trade, distance and oil production to reduce the influence of outliers and scale down the variance.Footnote ²⁰

Modelling

To test H1, in which the outcome is binary (support or no support), we follow common practice in the literature on third-party interventions (e.g. Kathman Reference Kathman2010; Koga Reference Koga2011; Regan Reference Regan1998) and use logit models with robust standard errors clustered on the dyad. In the online Appendix, we also show results using linear probability models with fixed effects in a monadic data structure as well as probit models. In testing H2a and H2b, we use multinomial logit models as the dependent variable has four separate outcomes: no support, troop support, weapons support and indirect support. Multinomial logit is a common estimation technique in studies on conflict outcomes (e.g. Salehyan et al. Reference Salehyan, Gleditsch and Cunningham2011) as well as third-party interventions (e.g. Koga Reference Koga2011). The estimates in the multinomial models show the effect of the predictor relative to the baseline outcome (no support).