US Foreign Aid, Interstate Rivalry & Incentives for Counterterrorism Cooperation

Transcription

1 US Foreign Aid, Interstate Rivalry & Incentives for Counterterrorism Cooperation Abstract A common strategy pursued by states targeted by international terrorist groups is to provide economic and military assistance to the states that host these groups. This is thought to increase their capacity or their willingness to crack down on the groups, but very little work to date has looked at whether this strategy actually leads to desirable outcomes. This article offers an explanation for why a strategy of foreign aid-for-counterterrorism can be successful in some contexts, but counterproductive in situations in which recipients strategic priorities are at odds with those of the donor. Specifically, I argue that host states receiving US foreign aid that are involved in an ongoing interstate rivalry will use the aid to arm against their rival, rather than to undertake counterterrorism. These states thus have incentive not to disarm terrorist groups, but rather to play-up the threat of terrorism in order to continue receiving aid concessions. Using data on US foreign aid and terrorist activity in recipient countries, I employ a series of duration and count models to demonstrate that while US foreign aid can help to decrease terrorist activity in states with no rivals, the opposite is true in states with at least one rival.

2 Introduction Since the 9/11 terrorist attacks, the United States has provided Pakistan with over $8 billion in total foreign assistance, including more than $2.4 billion in military aid. This assistance was intended to persuade the Pakistani government to combat the Taliban and other terrorist groups within Pakistan, and to bolster its capacity to do so. However, Pakistan has significantly different strategic priorities in South Asia than does the United States. The Pakistani government s primary concern since independence in 1947 has been hedging against its militarily more powerful neighbor and rival, India (Haqqani, 2005; Siddiqa-Agha, 2001). According to some observers, Pakistan has been using a sizeable portion of US military aid not to defeat terrorist groups, but rather to engage in nuclear and conventional arms buildups in anticipation of a possible future conflict with India (Rashid, 2008; Haqqani, 2005). The Pakistani intelligence service, the ISI, has also been accused of supporting the Haqqani network as a tool to counter Indian political influence in Afghanistan. Furthermore, others have speculated that the Pakistani regime fears the possibility of losing US support in the region if the threat from terrorism is reduced and the region becomes stable (Siddiqa, 2009). From the perspective of Pakistan, prolonging and amplifying the threat of terrorism within its borders has the potential to yield financial and strategic benefits into the future. Despite billions of dollars in aid over the past 15 years, Pakistan remains host to many terrorist groups and is still quite unstable (START, 2012). This may be due to the fact that much of the aid sent to Pakistan has not been put toward counterterrorism, police, and economic development as the US has desired, but rather toward achieving a more favorable balance of power with India (BBC, 2009). This is one example of what I demonstrate is a systematic trend of aid recipient states manipulating US interests in order to continue receiving aid. In this article, I offer an explanation for why this happens in some cases. Specifically, I argue that sending foreign assistance to a country for the purpose of combating terrorism signals a donor s strategic interest in the recipient. Because the donor s national interests are at stake, the recipient 2

3 is confident that aid will not be withdrawn even if, it suspects that a large portion of it is misspent. Here, I argue that a recipient that is part of an ongoing interstate rivalry with a third state will divert foreign assistance and other resources toward efforts to build capacity against its rival, rather than toward counterterrorism as the donor would prefer. An important contribution of this article will be the argument that in some cases, the presence of a domestic security threat in a country can be a strategic benefit for a government. This runs contrary to the more intuitive assumption that the dominant strategy for governments is always to decrease its level of terrorist activity and instability. Rather, I contend that large amounts of foreign aid and pressing strategic priorities (here, interstate rivalry) alter recipient incentives, making the maintaining a terrorist threat more beneficial to the host government than eliminating it completely. The following section summarizes previous attempts to assess the effectiveness of aid in eliciting cooperation. Next, I present a theory of counterterrorism aid beginning with the problem of aid fungibility, which is further compounded by interstate rivalry and informational asymmetries. In the third section, I test two implications from the argument empirically. Results indicate that higher levels of US foreign aid are associated with a significant increase in the number of anti-american terrorist attacks, as well as with longer terrorist campaigns in recipient countries. Aid to countries with no rivals decreases the length of terrorist campaigns, but has no effect on the number of attacks. Security motivations for aid allocation Several recent empirical studies in political science have examined the effectiveness of aid in eliciting cooperation from the recipient state. In his research on aid to Sub-Saharan African countries from 1975 to 1997, Dunning (2004) argues that due to the threat from the Soviet Union during the Cold War, the conditions placed on aid by Western donors had little credibility. African countries knew the strategic motives driving Western aid allocation, and 3

4 knew that the threat to withdraw aid if they did not comply was usually an empty one. As a result, they promised to implement democratic reforms, accepted the aid, and then reneged on their promises. Bearce & Tirone (2010) tell a similar credible commitment story to explain the effect of aid on economic performance in all recipient countries from 1965 to Using economic reforms and growth as their dependent variables, they show that levels of OECD aid had no effect during the Cold War, but a positive and significant impact on both reforms and growth after Tessman et al. (2011) use a scale of cooperative-conflictive actions to measure the effect of military aid on cooperation. Surprisingly, they find that higher levels of US military aid are associated with less cooperation from the recipient. They also find that uncooperative recipient actions tend to be followed by an increase in military aid in the following period, while periods of cooperation lead to aid cuts. Thus, for a recipient, non-cooperation pays, while cooperation is punished. Stone (2004) makes a similar argument about IMF lending in Africa. IMF program recipients who are strategically important to IMF lenders violate conditions more often, as they know that the lenders will intervene to prevent strict enforcement of conditions. Girod (2012) find that, even though conditionality may be unenforceable, recipients will comply with World Bank loan conditions if they are heavily dependent on foreign aid and do not have access to other sources of revenue. Finally, Bapat (2011) argues that US counterterrorism aid creates perverse incentives for host states to maintain terrorist activity at low levels so as to continue receiving assistance to combat it. At the same time, it is still rational for the US to continue giving aid in order to prevent the host government from collapsing or striking a deal with the terrorist group(s). The present study fits well within the framework of previous research on aid and recipient cooperation. In addition to the donor s credible commitment problem, my argument introduces rivalry to the analysis as a way to capture how recipient priorities condition foreign aid effectiveness. One additional contribution made by this article is that while the studies 4

5 discussed above use the credible commitment explanation to account for variation in democratic reforms (Dunning, 2004) and economic performance (Bearce & Tirone, 2010), with the exception of Bapat (2011), no attempt has yet been made to apply this framework to explain the effectiveness of aid for counterterrorism, a problem that is particularly relevant to the current international environment. This study builds upon Bapat s (2011) perverse incentives argument, but will argue that an aid recipient s strategic priorities in this case, interstate rivalry are important determinants of how effective foreign aid can be at eliciting counterterrorism cooperation. I also build upon extant research on the link between rivalry and terrorism. Both Conrad (2011) and Findley et al. (2012) find that terrorism occurs with greater frequency in rivalrous dyads; I show that this is not always the case, and that rivalry increases terrorism primarily when foreign aid is involved. My findings even suggest that states involved in rivalries tend to experience shorter terrorism campaigns. Additionally, past studies of foreign aid effectiveness have not given due consideration to recipient incentives and government counterterrorism behavior. This is a significant shortcoming, as foreign aid is often allocated with the understanding that the host government will cooperate toward achieving the donor s counterterrorism goals. With a couple of exceptions (i.e., Bandyopadhyay et al. (2010), Bapat (2011)), previous examinations of the aid-terrorism relationship have focused on the indirect effects of foreign aid on terrorism, either by asking how foreign aid decreases terrorism through increases in education Azam & Thelen (2008), or through allocations to other sectors, such as health and civil society (Young & Findley, 2011). While it is not my goal to find fault with these studies, I argue that a primary theoretical focus of any study of the effects of foreign aid on terrorism should be on the incentives and behavior of the recipient government. 5

6 A theory of rivalry & foreign aid Fungibility A substantial body of empirical research on the question of aid fungibility has concluded that most aid is fungible, and has pointed to this as a serious impediment to aid effectiveness. Aid fungibility, as defined by Hagen (2006), means that it is possible for the recipient to divert aid away from the donor s intended purpose. In an analysis on the effects of foreign aid on expenditure in the Dominican Republic, Pack & Pack (1993) finds evidence of substantial diversion of foreign assistance away from its intended purposes. Feyzioglu et al. (1998) actually find reductions in recipient spending in sectors for which aid is intended. Devarajan & Swaroop (2004) approach the problem from a slightly different angle, pointing out that the link between public spending and aid is not straightforward. Some aid may be targeted for specific programs in which the recipient government would have invested even in the absence of aid. In this case, foreign aid simply relaxes the recipient s budget constraints, freeing up funds to be used in other areas. This is likely to be the case for counterterrorism aid: the recipient government would probably have invested some of its own resources in combating violent groups anyway. As a result, the government may see counterterrorism aid as either money to be used for other projects or, at worst, as an opportunity for rent-seeking. For example, the US Embassy in Islamabad reports that at least half (and perhaps as much as 70%) of $5.4 billion in American aid to Pakistan in 2007 went unaccounted for (Rohde et al., 2007; Walsh, 2008). In addition, Pakistan has also allegedly used a substantial portion of this money to boost its military capacity to fight India, its longtime rival. Aid diversion and interstate rivalry Counterterrorism campaigns can require a vast commitment of political, military, and financial resources (Kilcullen, 2010). If the recipient has other, more pressing strategic imper- 6

7 atives, it will have little motivation to devote its own men and materiel to fight a costly war on behalf of another country at the expense of those other interests. Aid fungibility is unlikely to matter if the recipient views donor goals as a top priority (Hagen, 2006). In that case, the recipient will theoretically want to channel all of the aid received toward the sector for which it was intended by the donor. However, this is typically not the case, as recipients of US aid often have different priorities, such as the pursuit of interstate rivalries. Interstate rivalries are characterized by competition between two states who view one another as a security threat. Importantly, these states see one another as enemies not just in the moment, but into the foreseeable future as well (Thompson, 2001; Thompson & Dreyer, 2011). 1 Given this indefinite time horizon, rivals will want to build up their militaries now in order to hedge against power shifts and threats from their adversary in the future. Previous research finds that long-term rivalry dyads are disproportionately more likely to engage in arms races with one another (Rider et al., 2011; Rider, 2009). In the context of this study, states that are involved in a rivalry, and also receive large amounts of foreign aid, enjoy an advantage in that they are partially freed from having to raise taxes and manufacture equipment to finance their own military buildup. Foreign aid donors like the United States often supply equipment directly to the military, or provide the recipient with grants with which it can then purchase equipment. This enables the recipient state to engage in arms buildups while avoiding the negative consequences typically associated with increased military expenditures, such as rising deficits, inflationary pressures, decreases in investment capital, and unrest resulting from cuts in social spending (Rider, 2009). External rivals can also serve as a focal point for nationalist sentiments. Politicians can benefit politically by rallying constituents around a common, external threat that can also serve as a convenient scapegoat to distract attention from domestic problems. Thus, politi- 1 While some literature on rivalry has suggested that rivals power or military spending rather than the presence of rivalry alone are the most important determinants of rivalry behavior (Collier & Hoeffler, 2007;?), the rivalry coding I use in this article explicitly accounts for whether the states involved perceive one another as a threat and are competing with one another militarily. This variable is coded in part according to the spending patterns of rival states. 7

8 cians in recipient states are able to continue the politically expedient rivalry at little cost. On the other hand, a state with no rivals has less incentive to secure future aid concessions, and thus sees little value in prolonging a potentially destabilizing terrorist campaign simply to receive aid in the future. Donors may attempt to attach to aid packages conditions which must be met in order for aid to continue, but for reasons that I expand upon below, these conditions are often non-credible and tend to be violated with impunity. Donor commitment problems Donor states often use conditions on aid packages which the recipient must fulfill if it wishes to continue receiving aid. Such conditions often specify reforms to be undertaken or a level of effort that must be exerted, or simply that certain outcomes must be achieved, with the implicit threat that aid will be withdrawn or cut absent these reforms or outcomes. For the donor, this is the most attractive strategy to induce recipient cooperation, as it entails the fewest costs. Indeed, many believe that credible conditionality is the only way for aid to be effective at inducing policy change (Killick et al., 1998; Easterly, 2001; Dunning, 2004; Stone, 2004; Collier, 2007; Riddell, 2007; Bearce & Tirone, 2010). However, for conditionality to serve as an effective persuasive mechanism, threats to withdraw aid must be viewed by the recipient as credible. In other words, the recipient must fear an aid reduction if it does not comply with conditions set by the donor. In this section, I discuss why issuing credible threats to withhold strategic counterterrorism aid is quite problematic for the donor. Aid conditionality is often not credible for a number of reasons. First, for conditions to be credible, recipient non-compliance must be observable to the donor (Buchanan, 2000; Gibson et al., 2005). However, counterterrorism effort is often subtle and unobservable, and the information requirements for such monitoring are difficult to overcome without some violation of the recipient s sovereignty. An additional barrier to credible aid conditionality, and perhaps the most important for this article, is the strategic interest that prompted the aid in the first place. Suspending 8

9 counterterrorism aid might not be considered a viable option for donors unwilling to take that risk. American officials are often under considerable domestic political pressure to do something about threats to US interests. The public careers of politicians in the US may be at risk if voters or lobbies perceive the withdrawal of military and economic support for these countries as unwise in the face of threats to the US But because military action to counter these threats is rather costly, the US will almost always opt to delegate the fight by bolstering the military, economic, and political capacity of these states through the provision of aid (Siddiqa-Agha, 2001; United States Congress, 2008; Boutton & Carter, 2013). Thus, even if the US can verify that the host state is non-compliant, cutting aid may constitute an unacceptable level of risk for the donor state. Riddell (2007) observes that although conditions have been attached to almost all aid flows, aid is rarely withdrawn if the conditions are not met, and if it is, aid typically resumes shortly thereafter. Importantly, this observation was made about aid designed to bring about economic reforms rather than aid for security purposes. If punishment is so rarely enacted in the case of the former, it is even less likely in the latter case because the donor s commitment problem is intensified. As mentioned earlier, Tessman et al. (2011) find that defiant recipient states are often treated to aid increases following non-compliance, since the issue for which aid is given remains unresolved because of noncompliance. The United States and other donors have refrained from placing conditions on aid packages to strategically important countries for fear of creating a backlash and thus a more defiant partner. Specific conditions attached to aid packages are often viewed by recipients as violations of sovereignty, and sometimes create political upheaval in the recipient country (Hagen, 2004). In the wake of an uproar within the Pakistani elite over the initial terms of a recent US aid package, the United States softened the conditions contained in the 2009 Enhanced Partnership with Pakistan Act to state that Pakistan must simply show evidence that it is combating terrorism in order to continue receiving aid (Perlez, 2009; United States Congress, 2009). 9

10 The story I tell here is that when a donor such as the US sends foreign aid to a country in which violent threats to its interests are based, the recipient country s strategic priorities determine whether or not the aid will be effective in achieving counterterrorism cooperation. When the attention and resources of a recipient state are not focused on the pursuit of an interstate rivalry, foreign aid has the potential to be effective. However, when the recipient is part of a rivalry, the aid will be counterproductive for the reasons described above. Because of the donor s informational disadvantage and commitment problem, the recipient feels free to do as it pleases with the aid, and the donor has limited ability to prevent this. Hypotheses The argument outlined above points to expectations about the quality and/or quantity of counterterrorism effort undertaken by a recipient of US foreign aid. A direct test of these arguments would necessitate very detailed data on the counterterrorism behavior and capacities of US aid recipients. Unfortunately, the concept of counterterrorism remains ill-defined by scholars, spans across numerous policy areas, and can be quite difficult to observe. Counterterrorism effort could include overt activities such as military and police raids on terrorist hideouts; arrests; assassinations; or passage of anti-terrorism legislation. However, more covert, less observable actions, such as back-channel negotiations with a terrorist group; torture; or the freezing of financial assets could also be included, and thus any measure of counterterrorism that relies on observable indicators will be inevitably incomplete. Furthermore, observing a government s attempts to deliberately manipulate the level of terrorism present which is something that this article argues recipient with rivals are doing is very difficult. Perhaps jailbreaks would happen more frequently, as we have seen in Iraq, Yemen, and Pakistan recently (Reuters, 2013; Johnsen, 2013; Foreign Policy, 2013), although again, systematic data on these events is sparse. While systematic measures of these concepts are currently unavailable, some implications 10

11 of this argument can be evaluated empirically. The argument I discuss regarding aid diversion and interstate rivalry points to two implications about terrorism we should expect to see in states receiving foreign aid from the United States. Given a greater emphasis on counterterrorism in these states, we would expect terrorist groups not to last as long, or be more likely to fail, in a given year. However, I expect that host states involved in interstate rivalries will pursue the rivalry instead of US counterterrorism goals. In addition, these states will have incentive to play-up the level of terrorism within their borders in order to keep US foreign aid flowing, ensuring future resources with which to enhance their capacity vis-à-vis their rival. Therefore, we should expect that groups based in these states to last longer. First, I argue that without an interstate rivalry toward which to divert aid, recipient states are more likely to channel aid toward counterterrorism. In these cases, higher levels of foreign aid should lead to better counterterrorism outcomes for the donor. Therefore, groups in these states should be expected not to last as long. Hypothesis 1a. As foreign aid to state i increases, groups based in that state will be more likely to fail in year t if that state has no rivals. According to the argument, states receiving higher levels of foreign aid that are also involved in a rivalry should have incentive to maintain the groups as a prolonged source of leverage in order to continue receiving aid as a means to build capacity against its rival. Hypothesis 1b. As foreign aid to state i increases, groups based in that state will be less likely to fail in year t if that state is involved in a rivalry. The second set of hypotheses concerns how rivalry will condition the effect of foreign aid on the level of terrorist activity in recipient countries. It is reasonable to point out that the goal of this aid may not be to completely destroy the groups, but rather to enable the regime to prevent attacks against US citizens and interests. As long as its interests remain safe, the US may be indifferent to the prolonged presence of terrorist groups in recipient countries. 11

12 However, given their desire for continued US assistance to bolster their capabilities against their rival, recipient states may seek further assurance of future aid. Fearful that the United States would devote its aid resources elsewhere once it feels its interests are safe from attack, recipient regimes may turn a blind eye toward attacks against US interests, and may even engage in deliberate sponsorship and support of such activity (Byman, 2005). Therefore, in addition to longer-lasting groups, we might also expect more terrorist attacks against US interests in aid-receiving states with rivals. This leads to the following hypotheses: Hypothesis 2a. States that are not involved in an interstate rivalry will experience fewer anti-us attacks in a given year as foreign aid increases. Hypothesis 2b. States that are involved in an interstate rivalry will experience more anti-us attacks in a given year as foreign aid increases. In the following section, I outline the data, research design, and methodology used to assess these hypotheses. Data & research design Dependent variables To test my hypotheses about the effect of US aid on a recipient state s incentive to combat insurgent groups, I rely on two types of data. First, I use a modified version of the data used in the Jones & Libicki (2006) study on how terrorist groups end in order to test my hypothesis about group duration. For the second set of hypotheses in which terrorist attacks are the quantity of interest, I take data from the Global Terrorism Database (GTD). These variables are described in detail below. 12

13 Group duration The Jones & Libicki (2006) data identify 648 violent non-state groups that exist or have existed throughout the world between 1968 and Of these groups, 45 were based inside the United States and are thus not included in the present study. This leaves 603 internationally-hosted, non-state actors during this time period. I converted the data into binary time-series cross-sectional data, with one observation per year per country that the group was in existence. This type of arrangement is best to analyze the factors that cause a group to fail or endure, which is what I seek to explain here. In a few cases in the data, groups had legitimate bases in more than one country. When this was the case, I sought confirmation of multiple bases from US Congressional Research reports, the Global Terrorism Database (START, 2012), and from histories of the groups themselves when available. Where appropriate, I coded a group s presence in multiple countries from the date that these bases were established. These groups typically began operations in the multiple countries more-or-less simultaneously. For example, Jemaah Islamiyyah began operating in the Philippines in 1993, but quickly established bases in Indonesia and Malaysia. This group is thus coded in all three countries from Similarly, the Salafist Group for Call and Combat has carried out attacks in Morocco, Algeria, Mali, and Mauritania. I feel that this coding strategy is appropriate because the United States recognizes that these groups often traverse freely between countries, and its aid patterns reflect this (Ellis, 2004; Boutton & Carter, 2013). Because I am trying to explain action taken by the recipient state against these groups, my dependent variable is the failure of a group in year t due to actions taken by the state. The Jones and Libicki data code five different types of group failure: accomplishment of group goals; entry into politics; defeat by the military forces of an outside state; defeat by police action; and dissolution/splintering. Only the latter two are well-suited to my purposes, since these are the result of actions taken by the host state, which is what donors have in mind when providing foreign aid to these states. Therefore, I combine police failures with 13

14 splintering to form my dependent variable. This variable takes a value of 1 if the group fails in year t, and 0 otherwise: 0 if group is active in year t or ends for reasons unrelated to host state action y i,t = 1 if group fails in year t due to internal dissolution or police force When a group fails, it leaves the data and is not included in any subsequent years; there are 206 failures in the data. Attacks Data on terrorist attacks are taken from the Global Terrorism Database (START, 2012), which is useful as it contains information on the location, year, and primary target of over 80,000 terrorist attacks from 1970 to I construct three separate dependent variables based on the primary target of each attack: 1) a count of the number of attacks directed at US interests in a host country in a given year; 2) attacks directed at non-us targets; and 3) total attacks in a host country. An important point is that GTD classifies terrorist attacks according to the country in which they take place, and not according to the nationality or origin of the perpetrator(s) (START, 2012). For example, a cross-border attack on the US embassy in Nairobi by the Somali group al Shabaab would be associated with Kenya, rather than Somalia. This is appropriate for my purposes, since the US responds to terrorism primarily by allocating foreign aid to countries in which terrorist activity takes place (Boutton & Carter, 2013). Independent variables US foreign aid The theory outlined above specifies that levels of US foreign aid provided should influence the likelihood of a group failing in year t, conditional upon the presence or absence of interstate 14

15 rivalry in the recipient. I construct three separate explanatory aid variables, one each for levels of military, economic, Defense Department security aid, and total aid (the first three categories added together), 2 all of which are taken from the US Agency for International Development s Greenbook (United States Agency for International Development, 2013) and measured in constant 2009 US dollars. Defense Department security assistance is channeled through the Pentagon and is allocated with the explicit aim of bolstering military capacity of recipients and increasing their security cooperation with the United States. The Department of Defense began allocating security assistance to countries in I make several adjustments to these variables. First, raw aid variables are highly rightskewed. In my data, US aid ranges from $0 to over $17 billion, with a mean of $754 million. To correct for this, I take the natural log of (1 + aid), which smooths out the distribution of the aid variables by reining in outliers. Taking the natural log of aid also allows for the decreasing marginal returns to aid. Finally, I lag each aid variable by one year in order to account for the fact that aid received in year t is unlikely to affect the fate of a given group in year t. It is more likely that aid given to a state in year t 1 will be the relevant factor in the fate of a group in year t, as aid may take time to have an impact in the recipient country. I also estimate and present the results of models using aid normalized by population, or aid per capita. Interstate rivalry For interstate rivalry data, I use Thompson & Dreyer s (2011) work on strategic rivalries. Making use of information and descriptions contained in diplomatic and political histories, the authors classify states as rivals if key decision makers within both states regard each other as (a) competitors, (b) the source of actual or latent threats that pose some possibility of becoming militarized, and (c) enemies. (Thompson, 2001, 560). This approach to mea- 2 Importantly, the argument I make above applies equally to economic, military, and defense aid, since it has been shown that the presence of terrorism leads to increases in all types of US foreign aid (Boutton & Carter, 2013). Although I model them separately and then together as is common in past aid literature, I do not expect different effects across aid types. 15

16 suring rivalry differs in important ways from the dispute density approach favored by some (i.e., Goertz & Diehl (1993) and Klein et al. (2006), among others). The latter method classifies two states as rivals if they engage in a specified number of militarized disputes over a certain period of time. 3 However, while the presence of competition is necessary for a dyad to be called a rivalry in both measures, the dispute density approach by definition includes only highly militarized dyads. As Thompson (2001) points out, this can result in dyads being misclassified in a number of ways. First, two states may anticipate future conflict even if they have not yet engaged in a militarized dispute, meaning that the actual rivalry may begin well before the dyad crosses the dispute density threshold. Furthermore, the dispute density approach may be capturing cases in which major powers use military force often to coerce smaller powers. Such states are not likely to view one another as rivals, but would be classified as such by dispute density approaches. The strategic rivalry approach avoids this by limiting rivalries to dyads in which the states view one another as competitors. In short, I use the strategic rivalry approach taken by Thompson (2001) here because it is less biased toward rivalries that are already militarized, and because it is the most theoretically appropriate for this particular research question. 4 I use a dummy variable equal to 1 if a recipient state was involved in at least one interstate rivalry in a given year, and 0 if not. 5 To evaluate the hypothesis that the effect of aid on terrorist activity is conditional upon the presence of a rivalry, I multiply the rivalry indicator by each aid variable to create an interaction term. I also include a set of controls that reflect findings in extant quantitative terrorism literature. These include a natural log of GDP per capita (Maddison, 2012), natural log of country population (World Bank, 2010), democracy, military regime (Geddes et al., 2014), Cold War and Post-911 indicators, ongoing civil war indicator (Themner & Wallensteen, 2012), state 3 The exact dispute density threshold varies by researcher, but is commonly 6 disputes over a 20-year period. 4 Substantive findings do not change when I employ Klein, Goertz, and Diehl s (2006) rivalry measure. Results using this variable are reported in an online appendix. 5 Employing a count of the number of rivalries does not change the results. 16

17 sponsorship (Carter, 2012), the number of terrorist groups in country, group orientation variables (Jones & Libicki, 2006), and cubic time polynomials (Carter & Signorino, 2010). In the duration data, there are 6,424 group-years in the duration data across 92 countries and 603 groups between 1968 and In the fixed-effects models, 1,087 observations are dropped from cases in which there is no variation on the dependent variable, leaving me with 5,337 observations in the full models. The attack-count data contain 7,728 country-year observations, extending from 1970 to Endogeneity Aid cannot reasonably be believed to be exogenous; it is likely that the US allocates more aid to countries that experience more terrorist attacks, and that any estimate of the effect of aid on terrorist activity will suffer from bias as a result. Azam & Delacroix (2006) report a positive relationship between aid and terrorism not because aid has a perverse influence on terrorist activity, but because donor countries allocate more aid to countries that experience high levels of terrorism, and which are likely to continue experiencing elevated levels of terrorist activity in the future. Addressing the issue of endogeneity is therefore critical to conducting a proper empirical analysis. In the past, scholars have relied on instrumental variables and two-stage regressions in order to achieve a plausible claim of exogeneity (Sovey & Green, 2011). However, despite the difficulty of finding an instrument that satisfies the assumptions of IV estimation, Cohen & Easterly (2009) lament that the IV technique began to be used as a quick-fix by economists and political scientists to generate causal results. Rather than attempting to argue for an instrument for US foreign aid, I draw upon Azam & Delacroix (2006) and Young & Findley (2011) for two methods to address endogeneity. First, I estimate a series of linear regressions in which a set of exogenous explanatory variables are used to explain levels of economic, military, and total aid. The residuals from these 17

18 reduced-form equations are then included in the equations for the attack count models as a test for endogeneity bias. To explain levels of aid, I use the same regressors used in the attack count models, with the addition of a trade variable and a count of anti-us terrorist attacks. The second solution I use is to estimate a series of system GMM models accounting for endogenous regressors in dynamic panel data based on Arellano & Bond (1991). This method is designed for situations in which the number of time periods is small relative to the number of groups (countries, in this case), and in which the independent variables may not be exogenous. This technique has been used in foreign aid studies by Young & Findley (2011) and Dreher et al. (2008). The results for the main independent variables in the US attacks model do not change, and are reported in the online appendix along with a more detailed explanation of the system GMM approach. Results Aid, rivalry, & group duration To analyze the conditional effect of US aid on group duration, I estimate a series of grouped duration models in which the outcome variable is either failure by host state police force or internal dissolution (=1), or not (=0). The model is represented by the following equation: P (F ail = 1) = F (β 1 ln(aid) it 1 +β 2 Rivalry it 1 +β 3 ln(aid) it 1 Rivalry it 1 +β 4 X it 1 +γ i +ɛ), (1) where i indicates the country, t the year, F ( ) is the cdf of the logistic distribution, X is a vector of control variables, γ is the set of country fixed-effects, 6 and ɛ is an error term. All 6 Results of a Hausman test indicate that fixed-effects models outperform random effects models; the results of random effects estimation are not shown but are substantively the same. 18

19 time-varying independent variables are lagged. The estimation gives the effect of the independent variables on the probability of group failure in a given year, so a negative (positive) coefficient means that variable makes group failure less (more) likely. Further, the fixedeffects models estimate within-country variation to ensure that the results are not simply a reflection of cross-national variation. This is important, since the theoretical expectation is that as a country involved in a rivalry receives more aid, it will have increasingly less incentive to disarm terrorist groups. The results of these models are shown in Table I. 78 [Table I about here] Columns 2 4 contain the results of the logit models treating military, economic, and defense aid separately; estimates in column 5 are the results of the model grouping them together as total aid. The sign on the coefficient for GDP per capita is statistically significant in the expected direction in all four models: as a state grows wealthier, the likelihood of a group being eliminated in that country increases. Groups were more likely to fail at any given time during the Cold War, which is surprising given the norm of state sponsorship during this time (Byman, 2005). The estimates for total number of groups in a country are positive, meaning that as the number of other groups based in a country increases, group i is more likely to fail. It is not clear why this might be the case; one possible explanation is that a proliferation of groups in a country increases fighting between groups, causing them to die out or splinter more quickly. The population estimate is significant and positive in three of the five models, meaning that groups in more populous countries tend not to last as long. The coefficient for the democracy variable is negative and significant in the models containing observations only from 1993 and later, but is insignificant in all other models. The military regime coefficient is negative and significant only in the military and total aid models (columns 2 and 5). Consistent with past empirical research on sponsorship (e.g., Carter (2012)), groups that 7 Because the number of failures in the data are so few relative to the total number of observations, I also employed a rare-events estimator to correct for any possible bias; results are identical. 8 To save space, the coefficient estimates for the cubic polynomials of time are not shown in Table I. 19

20 are sponsored by outside states do not last significantly longer than groups that are not sponsored. None of the group orientation variables (territorial, regime change, or status quo) reach statistical significance in any of the models. Turning to my variables of interest, we see that the levels of US military, economic, Defense, and total aid are significant and positive, meaning that as aid to a country rises, a group based in that country is more likely to be defeated in year t. This coefficient should be interpreted as the effect of higher levels of aid when the recipient is not involved in a rivalry. The positive coefficients in all four models lend support to Hypothesis 1, which predicted that absent a rivalry, foreign aid should lead states to defeat terrorist groups more quickly. Interestingly, the estimated effect of rivalry when aid = 0 (i.e., the coefficient estimate on Rivalry t 1 ) is positive and significant in Models 1 and 4, meaning that groups based in recipient states receiving no aid and involved in a rivalry are more likely to fail in year t. This runs slightly counter to the intuition in Findley et al. (2012) and Conrad (2011), who find increased terrorist activity between interstate rivals. The other main hypothesis in this section concerns the effect of aid on group duration, conditional on the presence of rivalry. For this, we need to look at the coefficients on both aid and the aid rivalry interaction terms. To get the effect of aid conditional on rivalry, we need to add these coefficients (i.e., β 1 + β 3 from Equation (1))(Brambor et al., 2006). In this case, adding (-0.15) gives us -0.07, which we can interpret as the conditional effect of US foreign aid on group duration when the recipient has an interstate rival. Columns 6 8 contain results of duration models in which the natural log of aid per capita is used. The results particularly for the main independent variables do not change. Increases in aid per capita make group failure more likely in the absence of rivalry, and less likely when the recipient is involved in one or more rivalries. In fact, the conditional effect of aid in the total aid per capita model ( (-0.15) = 0.07) is identical to the conditional effect of total aid from Table 3. These results support both parts of Hypothesis 1. [Figure 1 about here] 20

21 Pr(Group Failure) Pr(Group Failure) Figure 1: Probability of group failure across values of military aid per capita Rivalry = 0 Rivalry = ln(military Aid per capita) ln(military Aid per capita) Figure 2: Distribution of Military aid per capita Since it is difficult to interpret the effects of these variables on Y simply by looking at the logit coefficients, Figure 2 displays the substantive effects of increases in military aid in the presence and absence of rivalry. The black lines connect the point estimates, while 21

22 the red lines represent the 95% confidence region around the estimates. On the left side of Figure 2, we see that when a host state is not involved in a rivalry, the probability of a terrorist campaign in that country is strictly increasing as it receives more military aid per capita. The opposite is true when the host is involved in an interstate rivalry. For instance, at $ military aid per capita (a value of 5 on the x-axis in Figure 2), the probability of group failure is 0.28 when the host is not involved in a rivalry, and 0.06 when it is. Again, these effects are large and statistically significant, and provide further support for Hypotheses 1a and 1b. Aid, rivalry, & terrorist attacks The results in the previous section suggest that US foreign aid may in some cases lead recipient governments to deliberately prolong terrorist campaigns. However, some could argue that the aim of US aid to these governments may not be the complete elimination of these groups, but rather to prevent US interests in the recipient country from being attacked. In order to maintain a level of threat, recipient states may seek either to facilitate attacks against US interests, or simply turn a blind eye toward such activity as a method to ensure the continued flow of US foreign aid to finance a rivalry. In this section, I present the results of the count models assessing Hypotheses 2a and 2b. [Table II about here] Columns 2 4 in Table II contain estimates from negative binomial regressions, with terrorism-relevant controls, in which the dependent variables are, respectively, total terrorist attacks in a country in a given year, attacks against non-us targets, and attacks against US targets. 9 In all three models, the coefficient on the aid variable is positive but not statistically significant, meaning that states receiving foreign aid which are not involved in a rivalry do 9 A number of robustness checks, including zero-inflated models and alternative specifications, are presented separately in an online appendix. The main substantive findings do not change. 22

23 not experience higher or lower levels of terrorism than those that are. 10 Rivalry alone is also insignificant; states involved in a rivalry but receiving no aid are not more or less susceptible to terrorism than those that receive aid. Turning to the other value of interest, the positive coefficient on the interaction term in the US attacks model (column 4) indicates that total US aid has a positive effect on anti-american attacks when the recipient is engaged in an interstate rivalry. This effect loses significance when total attacks and non-us attacks are modeled (columns 2 and 3), suggesting that recipient states may be turning a blind eye specifically toward terrorism directed at US interests in order to continue receiving US assistance. Hypothesis 2b is supported, but only as it relates to anti-us attacks; Hypothesis 2a does not receive support. 11 The Cold War and Post-911 indicators are negative, suggesting that terrorism against any target was less prevalent during the Cold War and after 2001 than during the 1990s. A country involved in a civil war is also likely to experience more terrorist attacks of both types. The effects of both population and GDP per capita are positive and large, which comports with previous findings in the terrorism literature. Democracies (as measured by Geddes et al. (2014)) do not experience significantly different amounts of terrorism in these models, but military regimes suffer from more terrorist attacks than other regime types. Military regimes are limited in their institutional capacity, and thus their ability to allow for the peaceful settlement of grievances, and rely heavily on coercion to quell dissent (Geddes, 2003). This finding also corroborates previous research on the link between regime type and terrorism (Conrad et al., 2014; Piazza & Wilson, 2013). The main results do not attenuate when the residuals from the aid regressions are included in the models, as indicated by the insignificant coefficient on endogeneity bias. This suggests that endogeneity is not a significant problem in the fixed-effects count models. 10 For purposes of presentation, I only present total aid results. 11 The online appendix presents the results of models estimated on the Cold War and post-911 years separately, showing that this relationship is consistent across time periods. 23

24 Conclusion This article has offered a possible explanation for why US foreign aid is effective at eliciting counterterrorism cooperation from some states, but may be associated with less compliance in other instances. Applying arguments from the foreign aid literature regarding preference divergence, I argue that when a recipient/host regime is engaged in a rivalry with another state, it will want to build up its capacity to hedge against future threats from its rival. Foreign aid provides these states with a way to build capacity at significantly less cost. Therefore, if a state with a rival is also receiving foreign aid from the United States to fight terrorism, it will have little incentive to decrease terrorism, but will rather play-up the threat in order to continue receiving aid windfalls with which to arm against its rival. The findings presented in this article have implications for the study of terrorism and foreign aid, a field which remains relatively underdeveloped. My findings suggest that while such aid may be effective in some cases, it may actually be achieving the opposite of its goal in other instances. A look at the top recipients of US foreign assistance can serve to corroborate these results. Terrorist campaigns in countries such as Israel, Pakistan, and the Philippines persist despite high levels of aid, yet the US continues and often increases its allocation to these countries. However, aid to countries such as Bangladesh and Indonesia has been relatively effective. These findings suggest that the United States should be more cautious when allocating aid to these countries, perhaps with more attention paid to the security priorities of the recipient. The results here open up several avenues for future inquiry. First, while these results are interesting and counterintuitive, they are only suggestive. More work needs to be done to specify and test causal mechanisms in the relationship between aid and counterterrorism effort. In order to test the argument about more directly, measures of this concept are needed. Second, information about the conditions attached to aid packages and perhaps more importantly, the credibility of those conditions would be useful. Although aid is fungible 24