For decades, the United States has relied heavily on the use of military assistance to train foreign security forces and to manage violence abroad. Yet these efforts have achieved, at best, a mixed record of success. In many cases, US military assistance appears instead to have intensified cycles of conflict, despite being meant to bring about stability.
Why is this the case? This question has puzzled policy makers for many years, and though scholars have recently begun addressing it (Bapat Reference Bapat2011; Biddle, MacDonald and Baker Reference Biddle, MacDonald and Baker2018; Ladwig Reference Ladwig2017), theories of military aid effectiveness have not systematically incorporated the politics and civil–military relations of the recipient country. This is an important oversight, and has left us with an incomplete understanding of why military assistance appears to uphold stability in some cases, and to undermine it in others. The balance of power between a leader, the security forces and other elites has important implications for political violence, and military assistance can – intentionally or not – disrupt this balance in favor of the incumbent regime. Thus as Brooks (Reference Brooks2019) has recently argued, the relationship between a leader and his ruling coalition and military – and how military aid alters that relationship – must be central to any theory of military aid effectiveness.
To address this shortcoming, I argue that the effects of military aid depend on the survival needs and priorities of the regime in the recipient country. These, in turn, are a function of both regime age and regime type. While the United States allocates military aid because it prefers a reliable government with apolitical, inclusive security forces capable of ensuring stability, the priority of a vulnerable leader is to consolidate power and coup-proof his regime. This process often generates – either directly or indirectly – violent opposition to the regime, and simultaneously decreases the security forces’ ability to manage such violence. Leaders are thus often careful when consolidating power. The receipt of external assistance, however, can create a moral hazard in which a weak leader feels emboldened to consolidate more aggressively, which can aggravate anti-regime violence. Using global data on regime characteristics and an instrumental variable approach to correct for the strategic distribution of US military aid, I show that military aid increases anti-regime violence when allocated to new regimes – particularly new democracies and military juntas – and to personalist regimes. By contrast, it has no effect on political violence in established, non-personalist regimes; new regimes that receive no military aid experience significantly less violence.
This article presents a novel argument to address a question of great importance to scholars of both international relations and comparative politics. Specifically, I identify a distinct mechanism through which military aid can increase domestic political violence. By describing how regime characteristics condition responses to external military support, it makes a significant contribution to our understanding of civil–military relations, aid and political violence. This theory of military assistance also makes sense of two anomalies often observed among recipients of US military aid. First, it explains why some US allies take actions that seem deliberately designed to intensify domestic conflict. Secondly, it accounts for why we often see US-trained militaries in fragile regimes collapse and retreat in the face of far smaller insurgent armies (for example, Goldstein and Mashal Reference Goldstein and Mashal2015). Military support, or the prospect thereof, provides cover for weak regimes to consolidate power and politicize their security forces – both of which provoke violence while degrading their ability to manage it.
In the next section, I review the small but rapidly evolving scholarly literature on the effects of military assistance, followed by a discussion of coup-proofing and its relationship to political violence. This is followed by a theoretical framework linking military assistance to political violence through a moral hazard triggered by regime vulnerability. Then, I introduce and describe the results of a series of quantitative analyses. I conclude with a discussion and suggest some policy implications and areas for future research.
Military aid effectiveness
US military aid typically entails military education and training programs to boost capacity and human capital (Jadoon Reference Jadoon2017; Savage and Caverley Reference Savage and Caverley2017), the provision of grants and weapons systems, logistical support and even in-country US military support. The goal of such assistance is to construct competent, inclusive, and apolitical security forces that are willing and able to support US policies and defend the recipient regime while upholding democratic norms (Atkinson Reference Atkinson2006; Shafer Reference Shafer1988, 92–95). Recent research, however, has noted that problems with military assistance programs arise when the priorities of recipient governments diverge significantly from those of the United States. Bapat (Reference Bapat2011) and Boutton (Reference Boutton2019b) argue that foreign aid can increase terrorism because, in some cases, it creates perverse incentives for recipients to sustain their terrorism problem in order to receive aid. Other arguments by Byman (Reference Byman2006), Watts, Shapiro and Brown (Reference Watts, Shapiro and Brown2007), Biddle, MacDonald and Baker (Reference Biddle, MacDonald and Baker2018), and Ladwig (Reference Ladwig2017) point to systematic preference divergence between the United States and its partner countries as the primary obstacle to military aid effectiveness. Principal–agent dynamics, they argue, prevent the donor from imposing conditions on or monitoring the actions of the host regime, allowing the host to pursue its own interests. As a result, only large-footprint nation-building operations are likely to result in successful military aid programs.
In line with recent research on military aid, I agree that donor–recipient preference divergence is a major cause of military aid ineffectiveness. I argue that this preference divergence is rooted in the political survival needs of the recipient regime, and that the goals of US military aid programs often directly conflict with those of a paranoid regime. The most urgent priority of these types of regimes is insulating themselves from the threat of a coup; military assistance can be useful not only for consolidating power, but also for surviving the often violent aftermath. The following sections discuss coup-proofing and its often violent repercussions before elaborating on how external military aid can exacerbate this relationship.
Coup-proofing and political survival in unstable regimes
In many well-established regimes, the likelihood of a coup is low, and thus leaders can govern without fear of irregular removal (Svolik Reference Svolik2009; Svolik Reference Svolik2015). Other regimes – such as newly established democracies and personalist regimes – are fragile or riven by factionalism, making leaders in these regimes vulnerable to removal by armed factions or other elites within the regime. The foremost priority of leaders in such regimes is to insulate themselves from the threat of removal and ensure both their hold on power and their physical survival.
Coup-proofingFootnote 1 can take a variety of forms. It can include reshuffling the military officer corps to prevent commanders from gaining too large a following (Quinlivan Reference Quinlivan1999); purging or excluding potentially disloyal officers or factions from the regime and military (Boutton Reference Boutton2019a; Roessler Reference Roessler2011; Sudduth Reference Sudduth2017b); military promotion based on political loyalty or ethnicity (Harkness Reference Harkness2016; Talmadge Reference Talmadge2015); the creation of overlapping and competing security institutions (DeBruin Reference DeBruin2018; Pilster and Böhmelt Reference Pilster and Böhmelt2011); or generally restricting access to state resources and the political process.
Coup-proofing and political violence
Although actions taken by coup-fearing leaders are meant to shield them from threats and solidify their hold on power, they often have a number of important effects which can exacerbate political violence while at the same time impairing the ability of the security forces to manage it.
First, power consolidation necessarily decreases the power of others within the ruling coalition who wish to both influence policy and constrain the leader from within the inner circle. This is particularly true in authoritarian regimes and new or fragile democracies. As Roessler (Reference Roessler2011), Sudduth (Reference Sudduth2017b) and Boutton (Reference Boutton2019a) explain, marginalizing certain factions creates disaffected group(s) with grievances against the regime. Roessler (Reference Roessler2011, Reference Roessler2016) shows that leaders' exclusion of factions from their governing coalitions has precipitated numerous African conflicts. Moreover, as we know from Crenshaw (Reference Crenshaw1981), elite marginalization and dissatisfaction is a primary driver of terrorism. This dynamic can be easily seen elsewhere as well. For instance, the ruling party in Bangladesh, the Awami League, recently purged from the armed forces a number of soldiers sympathetic to the opposition Bangladesh National Party and Jamaat-e-Islami. The purge has created animosity toward the ruling party within the military and society at large, which has benefited jihadist recruitment, including among marginalized soldiers within the military (International Crisis Group 2018, ii).
This also occurred in South Vietnam both prior to and during the war. According to a US foreign service officer, ‘Many of the original participants in the NLF [insurgency] had turned to it because they had been denied participation in South Vietnam's political process even in the role of loyal opposition’ (Chapman Reference Chapman2013, 189). Diem's resistance to power sharing assured that challenges to his regime would be violent. In fact, Hanoi's decision to establish the insurgency in the South was based largely on the volume of pleas they heard from marginalized southerners eager to take up arms. Similarly, the 1966 Buddhist uprising was the result of President Nguyen Cao Ky's purge of Nguyen Chanh Thi, a powerful Buddhist general whom Ky feared. The purge – which occurred just after Lyndon Johnson's pledge of renewed US military support at the 1966 Honolulu Conference – sparked a mutiny of Buddhist army units loyal to Thi, along with a campaign of violence from the Buddhist population that required Washington's assistance to suppress, and nearly toppled the regime (Sullivan Reference Sullivan2007).
Secondly, scholars are in broad agreement that coup-proofing negatively impacts the quality of the security forces and their ability to manage violence in a number of ways. Pilster and Böhmelt (Reference Pilster and Böhmelt2011) and Talmadge (Reference Talmadge2015) describe many of these in relation to conventional war, but they also apply to domestic security. First, the emphasis on political loyalty over merit in promotions necessarily degrades the competence and leadership of these units (Talmadge Reference Talmadge2015). In fact, coup-fearing leaders will often marginalize the most competent officers and instead reward incompetence in promotion. The most skilled commanders are also the ones most likely to develop an independent power base, and thus are the most capable of conspiring to threaten the regime. Such reshuffling also reduces morale among soldiers, who typically resent the installation of political loyalists in place of popular officers who rose through the ranks (Pilster and Böhmelt Reference Pilster and Böhmelt2011). This can create the perception among the rank and file that their career prospects are blocked, and that political loyalty is the only path to advancement. Soldiers who do not respect their officers, or who feel the regime is politicizing the military and promotion process, have fewer reasons to defend that regime, and may even take up arms against it. Talmadge (Reference Talmadge2015) notes several other ways in which threatened regimes undermine their militaries' operational effectiveness. These include limiting training exercises, centralization of command, frequent personnel rotation and compartmentalization of information to inhibit coordination.
Coup-fearing regimes often create bifurcated militaries in which certain units receive special treatment, while the remainder of the army is kept weak and under-resourced. Membership in the former is reserved for those who are ethnically or personally tied to the leader, or who have otherwise demonstrated political loyalty. These units are often headed by a relative or close ally of the leader, and are narrowly tasked with protecting the regime from domestic political opponents – sometimes including other factions of the military. In many cases, these units are the ones that receive training and equipment from the United States (Kedo and Goodman Reference Kedo and Goodman2015). The remaining rump of the military is left to wither on the vine, although Herbst (Reference Herbst2004) notes that many African regimes will train regular forces for external security duties such as peacekeeping, but break them up when they return home. This can intensify grievances and generate additional instability (Dwyer Reference Dwyer2018).
Roessler (Reference Roessler2016) describes yet another mechanism through which exclusionary coup-proofing can provoke violence. He says that coup-proofing – in addition to exacerbating grievances against the regime – also inhibits the regime's ability to effectively prevent and control rebellions by disrupting political networks. This mechanism echoes that of Greitens (Reference Greitens2016), who argues that the fragmented and sectarian security forces created by coup-fearing leaders can exacerbate political violence by impeding the regime's ability to gather intelligence about society, and increasing their propensity to use indiscriminate violence.
While coup-proofing may protect the leader from the threat of coups, it also has the dual effect of provoking anti-regime violence while simultaneously undermining the regime's ability to control it. Even if the coup threat is lowered, the risk of removal in a violent insurgency is still a distinct possibility. Vulnerable leaders would thus be short-sighted if they did not consider this when designing their power consolidation strategy. The provision of military assistance by an outside power in the event of domestic conflict, however, may alter these considerations, making it more attractive to use more exclusionary or inflammatory methods to consolidate power. The next section introduces the core of this moral hazard argument.
US military aid and moral hazard
As the discussion above illustrates, the coup-proofing and power-consolidation phase can be dangerous for the leader. The potentially destabilizing consequences therefore typically induce caution in leaders who wish to consolidate power. Svolik (Reference Svolik2012) notes that leaders maintain secrecy when attempting power grabs in order to avoid violent retaliation. Similarly, Sudduth (Reference Sudduth2017a, Reference Sudduth2017b) shows that dictators are prescient, preferring to purge their regimes when elites are disorganized or weak, and therefore less capable of inflicting retaliatory violence on the regime. After all, although more powerful regime figures pose the greatest threat to the leader from within the regime, these same figures are also the most capable of organizing large-scale, anti-regime violence if marginalized. Even insecure leaders are risk averse, and will avoid consolidating power when the risk of violence is high.
The receipt of external military support, however, may change the incumbent's perception of the potential costs of coup-proofing. If the leader anticipates that the United States will continue to provide military assistance to help counter a rebellion or terrorist campaign resulting from his consolidation of power, he may perceive the costs associated with coup-proofing to be lower. This may create a moral hazard in which the expectation of external military assistance causes the leader to consolidate power when he might otherwise refrain – such as when the likelihood of retaliatory violence is high – or to exercise less caution when doing so. This greater risk acceptance may manifest in more repression, ill-conceived purges of powerful figures capable of initiating a rebellion, or other forms of exclusionary behavior that precipitate violence. In fact, the expectation of military aid may make the leader more likely to consolidate power when the risk of violence is high.
Boutton (Reference Boutton2019a) shows in a recent study that dictators with external defense agreements purge their regimes more frequently, and when the likelihood of retaliatory violence is higher, resulting in more frequent civil conflict. Similarly, Song and Wright (Reference Song and Wright2018) argue that Kim Il-Sung was only able to consolidate personal control over his party and armed forces because of Soviet and Chinese military backing, which eased his fear of possible blowback that often accompanies power consolidation. Along similar lines, Cunningham (Reference Cunningham2016) argues that membership in the US security hierarchy gives client regimes a freer hand with which to repress dissent and prevent armed challenges. He shows that this governmental repression generates grievances, which increases domestic terrorism.
Moreover, coup-fearing leaders often choose units closest to them to receive US military training (Harmon Reference Harmon2014, 136–139; Kedo and Goodman Reference Kedo and Goodman2015). For instance, the United States spent nearly $600 million to train a variety of Yemeni special forces units to pursue al-Qaeda between 2001 and 2012 (Security Assistance Monitor 2017). Each of these was headed by a family member of President Ali Abdullah Saleh, and was used primarily against internal political opposition, rather than against al-Qaeda (Johnsen Reference Johnsen2013). Similarly, the Ugandan Special Forces Command has received US counterterrorism training, although they are also assigned primarily as personal protection for President Yoweri Museveni under the command of his son, Major General Muhoozi Kainerugaba (BBC 2017). A leader whose favorite security personnel receive US training may feel safer to engage in more aggressive coup-proofing tactics, as any resulting anti-regime violence can be countered by loyal, US-trained units.
Leaders in regimes that receive large amounts of military assistance may believe – rightly or wronglyFootnote 2 – that because of their strategic importance, such assistance is likely to continue. The US Army's counterinsurgency field manual states that to ease partner countries' fears of abandonment, ‘Constant reaffirmations of commitment, backed by deeds, can overcome that perception and bolster faith in the steadfastness of US support’ (Department of the Army 2006, 1–24). The fact that military aid is allocated according to US security interests creates the impression that the recipient regime is indispensable, and that additional aid would be forthcoming in the event of conflict. Recipient governments are fully aware of this leverage, and often seek to exploit it.
US support for Iraqi Prime Minister Nouri al-Maliki offers a good illustration of this moral hazard dynamic. The United States provided approximately $50 billion in military assistance between 2006 and 2014. The goal was to build the capacity of the security forces to protect the regime and enable Maliki to assemble a nationally representative governing coalition. Maliki had numerous opportunities to do this. His foremost fear, however, was being removed or killed by former Ba'athists; an inclusive government and military would have constituted an unacceptable risk. He therefore consolidated power along sectarian lines using textbook coup-proofing tactics (Dodge Reference Dodge2012). In fact, on the same day as the official US withdrawal in late 2011, special forces led by Maliki's son surrounded the houses of the vice president, deputy prime minister and minister of finance, all powerful Sunni officials. These actions provoked a violent reaction from parts of the Sunni population, ultimately culminating in the rebirth of the Islamic State insurgency. Maliki's coup-proofed security forces were unable and unwilling to risk their lives to defend the regime, and fled rather than defend major Iraqi cities from far smaller insurgent forces (DeBruin Reference DeBruin2014).
The violent response from factions within the Sunni population that he had marginalized was likely not unexpected. Maliki, however, was so confident that Washington would not abandon its investment in his regime – particularly after the United States intervened to back him following the 2010 elections – that he also believed Washington would continue to provide aid, and perhaps even intervene to fight the insurgency for him (PBS Frontline Reference Frontline2014; Khedery Reference Khedery2014). This led to a moral hazard: Maliki took US support as a blank check to consolidate power, betting that he would not have to bear the violent consequences alone.
Based on this argument, we should expect fragile regimes to take advantage of external military support and coup-proof more aggressively during their moments of weakness. We should also expect not to see increases in violence in vulnerable regimes that do not receive military assistance. Next, I discuss why we should expect to see these effects in newly established regimes – especially new democracies and military regimes – and in personalist regimes in general.
New Regimes
The establishment of a regime in power tends to increase over time. Huntington (Reference Huntington1968, 1) observed that the most important characteristic of any regime is its strength, of which age is a reliable proxy. Others also argue that healthy civil–military relations are partly a function of time (Kenwick Reference Kenwick2018; Ritter Reference Ritter2014). Figure 1, which displays the mean predicted probability of a successful coup across regimes at different temporal stages, supports this idea. The most insecure period in the lifespan of any regime is the first five years of its existence; the likelihood of a successful coup declines significantly in years 5 through 10 before levelling off in subsequent periods.
Figure 1. Coup success across regime type and regime age
Sudduth (Reference Sudduth2017a, Reference Sudduth2017b) shows that coup-proofing tends to occur just after a regime rises to power, arguing that elites are disorganized early on, and that they are less able to effectively mount a retaliatory coup against the leader during this initial phase. However, regimes are also quite vulnerable during the early period because the security forces may be incompetent, inadequately staffed or – most critically – loyal to the previous regime. This is true in both dictatorships and – perhaps especially – democracies. Leaders in newly established regimes must therefore create security forces that are competent, loyal and committed. This often involves reorganizing and purging the security forces quickly after gaining power. Figure 2 shows that the difference in the average rate of elite purges per year between new regimes (five years old or less) and more established regimes (older than five years) is more than 60 per cent.
Figure 2. Military purges in new versus established regimes
During transitions between discrete regimes, the new regime faces a period of uncertainty over issues such as policy reforms, the role of the military in the new regime, and ancien régime loyalists attempting to retake power (Keefer Reference Keefer2007; Mansfield and Pevehouse Reference Mansfield and Pevehouse2006). This is true in both autocracies – when one group of autocratic elites seizes power from another – and regimes transitioning from autocracy to democracy. New regimes are therefore weak and especially vulnerable to challenges during this initial period.
Consistent with research on authoritarian power consolidation (for example, Sudduth Reference Sudduth2017b; Svolik Reference Svolik2012), we should expect new regimes generally to be especially wary of their security forces. Rather than combating insurgents or terrorists, they will instead be focused primarily on coup-proofing, consolidating power and marginalizing potential threats to their own survival. While coup-proofing can be quite destabilizing, as noted above, the receipt of military aid will lower the expected costs of power consolidation if the leader anticipates that military aid will continue in the event of violent retaliation. This, in turn, may cause the leader to coup-proof more aggressively, increasing the likelihood of political violence in response:
Hypothesis 1
US military aid will increase political violence in new regimes.
Leaders in fragile regimes who do not anticipate military assistance may be left facing the potential violent aftermath of a power grab alone. They may therefore opt to behave as Svolik (Reference Svolik2012) or Sudduth (Reference Sudduth2017b) would expect and consolidate power with greater caution in an effort to minimize the risk of violence. Sudduth and Braithwaite (Reference Sudduth and Braithwaite2016) show that regime purges can reduce the risk of civil conflict in some cases. Such a leader might also opt for less exclusionary means, or to avoid consolidating altogether. Roessler and Ohls (Reference Roessler and Ohls2018) show that African leaders who believe that governing through exclusion will be too violent are more likely to form and abide by power-sharing agreements. Other inclusive forms of coup-proofing include the creation of institutions (Gandhi and Przeworski Reference Gandhi and Przeworski2007), parties (Geddes, Wright and Frantz Reference Geddes, Wright and Frantz2018; Magaloni Reference Magaloni2008), succession rules (Frantz and Stein Reference Frantz and Stein2017) and the distribution of patronage (Arriola Reference Arriola2009). We should thus expect new regimes that receive less military aid to experience less political violence as a result:
Hypothesis 2
New regimes that receive less military assistance will experience less political violence.
Regime Type
The effects of military assistance are likely to be heterogeneous across regime types. In this section, I consider fourFootnote 3 regime types and discuss theoretical expectations for each.
Democracies
The moral hazard is likely to be particularly acute in newly established democracies, wherein the coup risk is especially high (Svolik (Reference Svolik2008); also see Figure 1) and the legacy of the prior authoritarian regime is often strong. Extant research suggests that new democracies are particularly fragile, as they must contend with military or other autocratic elites who may be dissatisfied with their diminished role under democracy and wish to reclaim power (Cheibub Reference Cheibub2007; Keefer Reference Keefer2007; Svolik Reference Svolik2008). During the initial post-Suharto era between 1999 and 2004, for instance, the nascent Indonesian democratic government feared a return to military rule, and was thus reluctant delegate internal security matters to the military (Mietzner Reference Mietzner, Kunkler and Stepan2013). This resulted in abortive Indonesian counterterrorism efforts until 2005, by which point reforms had completely removed the military from politics.
Niger offers another recent example. President Mahamadou Issoufou survived a coup attempt in 2015, four years after Niger democratized in 2011. The president proceeded to purge nearly his entire senior officer corps – including the commanders of two elite counterterrorism units (Jeune Afrique Reference Jeune2017). While defense spending has risen substantially, much of the increase has gone to the president's personal guard (Bigot Reference Bigot2016). This has hampered US efforts to train the country's security forces. Although Niger is one of the top recipients of US security assistance in the region, the military as a whole remains ill-equipped, underpaid and unable to contain rising violence that is largely the result of the regime's own exclusionary policies (International Crisis Group 2017).
Hypothesis 3
US security aid will increase political violence in new democracies.
Personalist regimes
We should also expect military aid to exacerbate violence in personalist regimes, regardless of their age. Political power in these regimes is, by definition, limited to the leader and a small circle of regime elites; violent exclusion thus plays a central role. As Jackson and Rosberg (Reference Jackson and Rosberg1984, 421) put it, ‘the rivalries and struggles of powerful and willful men, rather than impersonal institutions, ideologies, public policies, or class interests, are fundamental in shaping political life’. Acemoglu, Robinson and Verdier (Reference Acemoglu, Robinson and Verdier2004, 162) note that a central feature of personalist rule is the presence of extractive institutions that enable ‘an individual, or a small group [to] use their power to transfer a large fraction of society's resources to themselves’. These types of predatory and exclusionary policies make it more likely that opposition to the regime will be violent; in fact, personalist leaders are far more prone to suffering violent removals throughout their time in power than others (Escribà-Folch Reference Escribà-Folch2013). Thus we might expect the moral hazard created by military aid to exacerbate such tendencies among personalist regimes throughout their tenure, not just at the outset.
Hypothesis 4
US security aid will increase political violence in personalist regimes, regardless of regime age.
Single-party regimes
The coup-proofing methods used in party-based regimes are significantly less coercive and exclusionary than those used in other types of regimes. Davenport (Reference Davenport2007) shows that, among autocracies, party-based regimes are the least reliant on repression to remain in power. Hegemonic parties are also adept at mobilizing citizens to vote in large numbers, paint the streets in party colors and demonstrate in support of the regime (Geddes, Wright and Frantz Reference Geddes, Wright and Frantz2018, 132; Magaloni Reference Magaloni2008, 729). The ability to draw upon a civilian support base is a critical coup-proofing mechanism. Military officers strongly prefer unopposed coups, and the presence of civilians supporting the regime in the streets can be a significant impediment to any attempts to remove it (Brooker Reference Brooker1995).
Strong ruling parties can deter military meddling, and help solve the commitment problem between the leader and other elites that often leads to exclusionary power consolidation and violence in other regimes. By making credible promises of future access to rents and powerful positions within the party apparatus for members, party-based regimes are able to inspire loyalty, co-opt potential opposition and prevent elite defections, all without resorting to coercion (Smith Reference Smith2005; Magaloni Reference Magaloni2008). This is true even in the early years of a single-party dictatorship: in all cases of single-party dictatorship – with the sole exception of Mexico between 1929 and 2000 – the dominant party was well organized prior to gaining power (Geddes, Wright and Frantz Reference Geddes, Wright and Frantz2018, 116). Thus, because these regimes rely less on violence to consolidate and retain power in the first place, I do not expect military aid to increase political violence in single-party autocracies.
Hypothesis 5
US security aid will have no effect on political violence in party-based regimes, regardless of regime age.
Military regimes
Military juntas rely more heavily on violence to remain in power than other regime types, particularly during the consolidation phase (Davenport Reference Davenport2007; Escribà-Folch and Wright Reference Escribà-Folch and Wright2010). In fact, destroying the opposition was often their raison d’être during the Cold War (Shafer Reference Shafer1988). In the wake of the purges of Muslim Brotherhood sympathizers from the regime after the 2013 military takeover, Egypt – consistently a top recipient of US military aid – has experienced a surge in violence. Al-Naba, the Islamic State weekly magazine, documents groups of Egyptian soldiers purged from the military after the coup because they were sympathetic to the Muslim Brotherhood. They have since joined the Sinai branch of the Islamic State as military commanders and are now organizing violence against the regime (al-Naba 2018, 9–10).
Whether the Egyptian regime would have perpetrated such violence against the Muslim Brotherhood had they not been a close US ally is unclear. Sudduth (Reference Sudduth2017b) demonstrates empirically that leaders who come to power via coups – which is true of all military juntas – are far more prone to violent power consolidation in the early stages than are leaders who come to power in other ways. This effect decreases, however, the longer a regime's tenure. We should expect the moral hazard created by external military support to exacerbate this tendency, leading to more political violence in newly established military regimes.
Hypothesis 6
US security aid will increase political violence in newly established military juntas.
Empirical analysis
Domestic Attacks
The theory predicts that security assistance to new regimes will encourage behaviors by the recipient government that provoke and facilitate anti-regime violence. I therefore use Domestic Attacks against the Host Country as my dependent variable, as this captures the perverse effects of US security aid on the stability of the recipient country. Data on domestic attacks are taken from the Global Terrorism Database (START 2018), which contains information on the location, year and primary target of over 80,000 terrorist attacks from 1970 onward. I classify each attack according to whether its primary or secondary target was the host country.Footnote 4
US Military Aid
The theoretical argument above points to the potentially adverse effects of security-related assistance. By better equipping and training recipient security forces, military assistance gives the recipient leader the confidence that he can consolidate power more aggressively and survive the fallout. Data on military assistance are only available for the United States, and were collected from the Security Assistance Monitor (2017), which reports the annual dollar amount spent on each program in each country. The database aggregates a variety of security-related grant and training programs, such as International Military Training and Assistance and Section 1206 and 1207 Defense Department assistance, and programs such as the Coalition Support Fund, Counterterrorism Training Fellowship Program or foreign military financing, which consists of grants to foreign governments to spend on their militaries. It also includes Department of State anti-terrorism assistance. I take the natural log of the three-year moving average 
$\left( \lpar{\displaystyle{{{\rm Ai}{\rm d}_{t-1} + {\rm Ai}{\rm d}_{t-2} + {\rm Ai}{\rm d}_{t-3}}\rpar / 3}} \right)$ of this variable to better capture the amount of aid a recipient is conditioned to expect.
Regime Age
Regimes are defined as a set of formal and/or informal rules for choosing leaders and policies. Each regime constitutes a different societal group from which leaders can be chosen. Using definitions in Geddes, Wright and Frantz (Reference Geddes, Wright and Frantz2014), I calculate the age of each discrete regime. I classify any regime less than six years old as new; prior studies on regime transitions have found this cutoff to be analytically useful (for example, Cook and Savun Reference Cook and Savun2016; Mansfield and Pevehouse Reference Mansfield and Pevehouse2006).Footnote 5 Figure 1, which displays estimates for successful coup probabilities in regimes at various stages of development, more or less validates this distinction. These estimates suggest that in most cases, the first five-year period is significantly more coup-prone than subsequent periods. After the first five years, coup probability drops precipitously, but remains relatively stable as regimes age beyond five years.
Regime Types
I also draw upon Geddes, Wright and Frantz (Reference Geddes, Wright and Frantz2014) to classify democratic and authoritarian regime types. This classification does a particularly good job of coding each regime based on the features that are theoretically important for this argument (see their codebook for details on the coding procedure).
Other independent variables
Each model includes a variety of additional covariates thought to influence levels of terrorism. These include GDP per capita, population, press censorship and human rights measures, and binary indicators of interstate rivalry, ongoing civil war, election years and coup occurrence.Footnote 6
Estimation
OLS and negative binomial estimation
I first test my hypotheses using both ordinary least squares (OLS) regression and negative binomial count models. I use ln(1 + Domestic Attacks) as the dependent variable in the OLS estimation, while I leave the raw count of attacks as the outcome in the negative binomial. All estimations include year- and country-fixed effects to control for unobserved time and unit-level heterogeneity. Models 1 and 3 contain the results of the count and OLS estimation, respectively, without the Military Aid × New Regime interactions. Coefficient estimates for βmil. aid – the effects of military aid in all regimes – are positive but insignificant in both models.
Models 2 and 4 add the New Regime × US Military Aid interaction terms. Estimates for the βmil. aid×new regime interaction are positive and significant, while βmil. aid – the marginal effect of military aid in established regimes – is small and statistically insignificant. As shown in the highlighted row near the bottom of Table 1, the estimated marginal effect of military aid in new regimes (βmil. aid + βmil. aid×new regime) is positive and statistically significant in both the OLS and count models.
Table 1. US military aid and domestic terrorism
Note: country and year fixed effects included in all models but not presented. Clustered standard errors in parentheses. +p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001
The estimates for New Regimes are positive in the non-interaction models: perhaps unsurprisingly, new regimes are, on average, more unstable than established ones. When the interaction terms are included, however, this effect is reversed (although the negative estimate is insignificant). This means that new regimes that receive no US military assistance do not experience significantly more domestic terrorism than other regimes. This casts doubt upon the possibility that the results are driven by higher levels of instability and violence in new regimes generally.
2SLS instrumental variable estimation
The analyses above may suffer from endogeneity resulting from unmodeled strategic behavior. If US policy makers allocate aid in response to instability, or if incumbents in new regimes seek external assistance to use against domestic opposition, the estimated relationship between military aid, new regimes and violence would be artificially biased upward. US military assistance to Afghanistan, Pakistan and Yemen are good examples of this (Boutton and Carter Reference Boutton and Carter2014). Alternatively, US officials may curtail the flow of military assistance to unstable regimes or those that are more likely to repress political opponents. For instance, the United States suspended military aid to Indonesia due to human rights concerns during that country's conflicts in East Timor and Aceh. Likewise, Tankel (Reference Tankel2018) notes that the United States dramatically reduced its security assistance to Mali in 2012 after the onset of the Tuareg rebellion. In these cases, findings would be biased toward zero.
Research on aid effectiveness has long been plagued by such problems. Recent techniques for addressing this concern in the cross-national context involve using time-varying characteristics or conditions within the donor country that may be correlated with aid allocation but plausibly exogenous to conditions in the recipient country. This characteristic is then interacted with the likelihood that a recipient will receive aid, which provides a cross-sectional component. Examples of this include Nunn and Qian (Reference Nunn and Qian2014), who use weather-induced variation in US wheat production to estimate the effects of US food aid on conflict. Dreher and Langlotz (Reference Dreher and Langlotz2017) exploit the correlation between donor government fractionalization and aid budgets to estimate the effect of development aid on growth.
To construct my instrument, I rely on a modified version of the approach used by Ahmed (Reference Ahmed2016). This method uses time-varying fragmentation within the US House of Representatives, which is equivalent to the annual difference in the number of Republicans and Democrats seated in the lower house of Congress. Prior research (for example, Alesina and Tabellini Reference Alesina and Tabellini1990) finds that more fragmented legislatures tend to produce larger budgets, and empirical patterns presented in the online appendix demonstrate that annual US military aid allotments correlate positively with fragmentation in Congress. The composition of the House changes on a biennial basis, and individual legislators are elected based primarily on district- or national-level political and economic conditions. Politics and violence in foreign countries are unlikely to be connected to election outcomes in the US House of Representatives. This variable can therefore be considered a plausibly exogenous source of variation in US military aid disbursements.
House fragmentation captures variation in aid disbursements over time only. To add cross-sectional information, I interact fragmentation with the average probability of a country receiving US military aid 
$(\overline {Pr{({\rm mil}{\rm.} \;{\rm aid}\; \gt \;0)}_i} )$, similar to past research on the effects of economic development aid (for example, Ahmed Reference Ahmed2016; Dreher and Langlotz Reference Dreher and Langlotz2017; Nunn and Qian Reference Nunn and Qian2014). Thus the time-varying exogenous shock – captured by House fragmentation – will be weighted most heavily in frequent military aid recipients such as Colombia, Greece and Pakistan, and less in countries that rarely receive military aid, such as Australia or Norway. The cross-sectional probability of receiving US military aid is highly correlated with the amount of aid received (see online appendix), but contemporaneous political behavior does not determine this probability, which is time invariant. Other channels through which this may impact violence are captured through control variables and unit-fixed effects, which I include in all models.Footnote 7 I also take the three-year moving average of legislative fragmentation.
The 2SLS instrumental variable model can be represented by the following:
$${\rm Domestic}\;{\rm attack}{\rm s}_{it} = \alpha _0 + \widehat{{{\rm Mil}{\rm.} \;{\rm aid}}}_{it-1:3} + X_{it-1} + \gamma _i + \xi _t + \epsilon _{it1}$$
$${\rm U}{\rm. S}{\rm.} \;{\rm mil}{\rm.} \;{\rm ai}{\rm d}_{it} = \alpha _0 + Z_{it-1:3} + X_{it-1} + \gamma _i + \xi _t + \epsilon _{it2}$$where Z it−1:3 is the excluded instrument 
${House \,fragmentatio}{\rm n}_{t-1:3} \times \overline {Pr{({mil}{\rm.} \;{aid})}_i} $, X it is a set of control variables, γ i and ξ t are cross-sectional and time-fixed effects, respectively, and ε it is a disturbance term. To account for potential serial correlation, I cluster standard errors by regime.Footnote 8 Columns 5 and 6 of Table 1 contain the results of the two-stage least-squares (2SLS) estimation without the interaction term, in which the excluded instrument predicts the endogenous regressor, ln(US military aid). The F-statistic is 88.24, well above the 10 per cent weak ID critical value of 16.38, indicating a sufficiently strong instrument (Stock and Yogo Reference Stock, Yogo, Andrews and Stock2005). As in the OLS and count models, the estimate for βmil. aid – which gives the average marginal effect of military aid across all regimes – is positive but statistically indistinguishable from zero.
Columns 7–9 contain the results of the 2SLS with the interaction term. Because there are two endogenous regressors – military aid and the Military Aid × New Regime interaction – Columns 7 and 8 contain the two first-stage equations. Again, the F-statistic is larger than the weak ID critical value, suggesting that the instrument is strong. The estimate for βmil. aid remains positive but statistically insignificant, meaning that the marginal effect of US military aid on domestic terrorism in established regimes (that is, those older than five years) is essentially zero. The estimated marginal effect of US military aid in new regimes – given by βmil. aid + βmil. aid×new regime – is positive and statistically significant. Note also that the coefficient for new regimes in which military aid is equal to zero – βnew regime – is negative, meaning that new regimes that do not receive military aid experience about 70 per cent less domestic terrorism. This provides additional evidence to support the central argument.
Figure 3 displays the marginal effect of increases in military aid.Footnote 9 In established regimes, the effect is small and statistically insignificant, while in new regimes it is positive and significant.
Figure 3. Marginal effects of military aid on domestic terrorism (OLS estimates)
Military aid, regime type and regime age: 2SLS estimation
To evaluate the hypotheses that military aid will have heterogeneous effects across regime contexts, I replicate the 2SLS interaction analyses above on subsamples of each regime type. The advantage of this approach is that it can demonstrate the strength of the instrument within each regime subsample, although the much smaller N in some subsamples may depress the F-statistic. These results are presented in Table 2.Footnote 10 In all cases except the military junta subsample, the F-statistic surpasses the weak ID critical value indicating instrument strength, likely due to the presence of only 192 observations in this category.
Table 2. 2SLS IV interaction models by regime type
Note: clustered standard errors in parentheses. Year and region fixed effects included in all models but not shown. First-stage results reported in the online appendix. +p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001
Column 1 shows that in established democracies, the marginal effect of military aid (βmil. aid) is small and statistically insignificant. The estimated marginal effect in new democracies – βmil. aid + βmil. aid×new regime – is 0.198, which is statistically significant at the p<0.01 level, consistent with expectations.
Column 2 shows that while security assistance has a positive but insignificant effect in new personalist regimes, it has a positive and significant effect in established personalist regimes. This is contrary to the intuition that military aid should increase political violence in new regimes, but it is consistent with past research on the effects of foreign aid on terrorism in personalist regimes (for example, Boutton Reference Boutton2019b). This may be because one of the key traits of new regimes in this argument – the desire to marginalize challengers – is characteristic of personalist regimes throughout their tenure.
Long-ruling Cameroonian president Paul Biya is a good example of the tendency of personalist leaders to circle the wagons and fight rather than negotiate when confronted with opposition (Shurkin Reference Shurkin2018). Biya may have been emboldened by US military assistance to deploy the US-trained Rapid Intervention Battalion against protestors rather than negotiate, thereby escalating the crisis into a civil conflict. Determining the precise nature of the mechanism, however, must be left for future research.
Military assistance appears to have no significant effect in party-based regimes, regardless of regime age. While the coefficient estimate on the interaction term in the junta subsample is positive and significant, the marginal effect of military aid in new military juntas is positive but does not reach the threshold for statistical significance.
Figure 4 displays marginal effectsFootnote 11 in democracies and personalist regimes by regime age. Military aid is estimated to have no effect on violence in established democracies, but a positive and significant effect in recently established ones. By contrast, the plot to the right shows the positive effect of aid on political violence in established personalist regimes. While the effect is also positive in new personalist regimes, the 95 per cent confidence interval includes zero.
Figure 4. Marginal effects of military aid on domestic terrorism in democracies and personalist regimes (OLS estimates)
Conclusion
The goal of this article has been to investigate the effects of US military assistance on political violence in various contexts, and to explain why we observe large disparities across cases. I find strong evidence that US military aid increases domestic political violence in newly established regimes – especially new democracies and military juntas – and in personalist regimes generally. It is unrelated to political violence in established, non-personalist regimes. This is true even when correcting for the strategic distribution of US military assistance using an instrumental variable. I argue that this is due to a moral hazard induced by external support and regime vulnerability.
These findings are important and point to several areas worthy of future scholarly efforts in both comparative politics and international relations. The argument that military aid alters leader incentives should lead us to ask how military aid influences leaders' decisions during periods of violent unrest, or when faced with nonviolent or electoral challenges. Will they be more inclined to use violent repression, or less likely to negotiate or share power? Likewise, more direct investigations into how the military organizational practices of a regime change in response to US military assistance would be useful. For instance, once could ask how military aid influences the type of coup-proofing strategies employed by a leader.
These results also raise questions about how external military assistance alters the proclivity of recipient governments to provide governance, public goods, and services to the population more broadly. For instance, Mali is a fragile democracy in which successive governments have neglected to address communal conflicts over land pasturage and water. This has created space for a major escalation of the conflict in central Mali, as jihadist groups arrive to these communities to find local conflicts ripe for exploitation (Benjaminsen and Ba Reference Benjaminsen and Ba2018). Tahir (Reference Tahir2017) points to a similar process in Pakistan. It is an open question whether, without the promise of external military assistance, such governments would have addressed domestic political, economic and social issues with greater urgency, thus denying space in which extremism can flourish. Given how little we know about how military aid affects the politics of recipient states, these are promising avenues for additional research.
The evidence that external military support can inadvertently destabilize allies also has stark implications for US foreign policy. Officials in the United States often speak of the need to support new or transitioning regimes – particularly democracies – with military aid or other forms of assistance to facilitate the establishment of order and the consolidation of democratic government. These results show that such assistance can in fact be quite destabilizing. New regimes that receive no military assistance experience much less political violence than those that do. These findings validate the concerns of Tilly (Reference Tilly, Evans, Rueschemeyer and Skocpol1985) and Herbst (Reference Herbst1997), who warned that external military support can distort elites' incentives and hinder the natural political development of recipient countries. The findings should force a reconsideration of how best to assist weak states with underdeveloped institutions.
Acknowledgements
I am grateful to Konstantin Ash, Whan Choi, Mark Crescenzi, Mike Kenwick, Joe Wright, and audiences at APSA and Peace Science for insightful comments on earlier drafts, and also to Töbias Bohmelt for providing helpful guidance during the review process.
Supplementary material
Data replication sets are available in Harvard Dataverse at: https://doi.org/10.7910/DVN/GOGH2L and online appendices are available at https://doi.org/10.1017/S000712341900022X.
