‘Pay for It Heavily’: Does U.S. Support for Israel Lead to Anti-American Terrorism?

ABSTRACT After the 9/11 attacks in particular, there has been a controversial discussion in the academic and public arena on whether the United States’ close relationship with Israel has made it a likelier target of transnational terrorism. Indeed, foreign terrorist organizations with various ideological profiles have repeatedly justified attacks against U.S. interests as punishment for the (purported) special relationship between the United States and Israel. We analyze the effect of various measures of U.S. support for Israel (e.g. U.S. military assistance to Israel) on anti-American terrorism for the period 1970–2014. Using both time-series and panel approaches, we do not find that more U.S. support for Israel systematically translates into more anti-American terrorism. Rather, other systemic (e.g. U.S. dominance in the international system) and local conditions (e.g. local state failure) are found to predict the patterns of anti-American terrorism. However, as a qualification to these general findings, we also provide some (preliminary) evidence that for terrorism originating from the Middle East and Northern Africa a favorable U.S. policy stance towards Israel may indeed contribute to more anti-American terrorism.


Introduction
A popular claim about transnational terrorism 1 directed against the United States, especially with respect to the 9/11 attacks, is that it is rooted in U.S. support for the state of Israel. For instance, a representative survey among U.S. citizens in the aftermath of the 9/11 attacks found that 85 percent of respondents believed U.S. support for Israel was at least a somewhat important cause of the attacks, while all other factors (e.g. Islam, lack of economic aid) were considered less influential (Davis and Silver 2004).
Indeed, U.S. support for Israel is cited as a motivating factor of Al Qaeda's anti-American terrorist activities. For instance, in his 'Letter to America ' Bin Laden (2002) writes that "[the] creation and continuation of Israel is one of the greatest crimes, and you [i.e. the Americans] are the leaders of its criminals. And of course there is no need to explain and prove the degree of American support for Israel. The creation of Israel is a crime which must be erased. Each and every person whose hands have become polluted in the contribution towards this crime must pay its price, and pay for it heavily." balance of power between the attacking terrorist organization and Israel in favor of the terrorist group, also making terrorist success (the long-run terrorist goal) more likely.
Indeed, there is some evidence for the strategic logic channel. Plümper and Neumayer (2010) find that military alliances (say, between the U.S. and Israel) expose nationals to an increased risk of terrorism originating in their ally's home country, where this effect becomes more pronounced as the power asymmetry between allies increases. Similarly, Neumayer and Plümper (2011), Gries, Redlin, and Meierrieks (2015) and Krieger and Meierrieks (2015) find that countries that receive more military support (e.g. in the form of arms exports) from the United States are more likely to generate anti-American terrorism.
Considering the nexus between U.S. support for Israel and anti-American terrorism, the strategic logic channel is likely to apply to two kinds of terrorist groups. First, to Arab-Palestinian groups that directly oppose Israel. As LaFree, Yang, and Crenshaw (2009) find, a number of these groups (e.g. the Popular Front for the Liberation of Palestine) have indeed repeatedly also attacked U.S. targets. Second, radical Islamist groups are generally sympathetic to the Palestinian cause as well as interested in liberating one of Islam's most holy sites (the Al-Aqsa Mosque) and in destroying the 'world jewry' as an enemy of Islam. 6 Again, such groups (e.g. Hezbollah, al-Gama'at al-Islamiyya) have also attacked U.S. interests (LaFree, Yang, and Crenshaw 2009).

The Support Channel
Besides terrorist organizations' goal of weakening their enemy, Plümper and Neumayer (2010, 77) name another important intermediate goal of these organizations: survival. To ensure survival, terrorist groups depend on support. This support may come in the form of manpower, funding, training, provision of matériel and sanctuary as well as intelligence (Paul 2010). Support for terrorist groups can come from various sources: sympathetic local communities and diasporas (popular support), states (state sponsorship of terrorism) and other terrorist groups. More support for terrorist groups from external sources is expected to facilitate terrorism. For instance, with external support it ought to become easier to evade capture (e.g. thanks to access to 'safe havens'), lowering the costs of carrying out terrorism.
Importantly, carrying out anti-American terrorism in the name of (perceived) U.S. support for Israel may allow terrorist organizations to tap into various sources of support. First, the U.S. tends to be an unpopular out-group in the Middle East, especially given its foreign policy behavior (Haddad and Khasan 2002;Tessler and Robbins 2007). While attacking an out-group always ought to be more popular than attacking fellow citizens, punishing the U.S. for its foreign policy failures may be a beneficial strategy to attract popular support especially for Arab-Palestinian groups as well as radical Islamist groups.
Second, anti-American terrorism may make terrorist groupsof all ideological stripesattractive allies for states that are hostile towards Israel for strategic or ideological reasons. These countries have an incentive to weaken Israel; they may achieve this by sponsoring anti-American terrorist groups so as to coerce the U.S. into reducing its support to Israel or, more generally, take a more isolationist foreign policy position that would consequently also allow for revisionist policies directed against Israel. Indeed, countries that have been hostile to Israel in the past have also supported terrorist groups that have attacked U.S. interests. Examples include the Philippine Moro National Liberation Front and Peruvian Tupac Amaru Revolutionary Movement (sponsored by Libya), Hezbollah (sponsored by Syria and Iran) and the Abu Nidal Organization (sponsored by Syria and Iraq). 7 Finally, terrorist organizations may gain support by allying themselves with other terrorist organizations. For example, through cooperation groups may share expertise, equipment and intelligence, thus aiding the terrorist groups' survival. Especially during the Cold War, Latin American and Western left-wing terrorist groups cooperated with Arab-Palestinian organizations (Shughart 2006). For instance, the Popular Front for the Liberation of Palestine at times cooperated with the Colombian Fuerzas Armadas Revolucionarias de Colombia and the German Red Army Faction. For Arab-Palestinian groupsin line with the strategic logic channel outlined abovesuch cooperation ought to have been beneficial as left-wing terrorist groups would regularly attack the United States (the common enemy of Arab-Palestinian and left-wing terrorism), thus diverting U.S. resources away from the Middle East and possibly also contributing to a more reluctant U.S. foreign policy stance. We can thus expect that increased U.S. support for Israel would have created additional incentives for cooperation between terrorist groups. In fact, the cooperation between Arab-Palestinian and left-wing terrorist groups was at times so close that the (less powerful) left-wing terrorist groups would not only target U.S. interests but also consciously adopt their ally's ideology and/or target its main enemy (i.e. Israeli citizens). For instance, the Japanese Red Army, originally a left-wing group, was closely allied with the Popular Front for the Liberation of Palestine as of the early 1970s and, in the following, adopted the latter's anti-Israel agenda. Subsequently, in the 1972 attack on Tel Aviv's Lod Airport, members of the Japanese Red Army killed or injured over 100 individuals on behalf of the Popular Front for the Liberation of Palestine (Shughart 2006, 23).

Hypothesis
In sum, the link from the U.S. support for Israel to anti-American terrorism may run via the strategic logic and/or the support channel. In the following sections we empirically test whether such linkages indeed exist, expecting support for the following hypothesis.
Hypothesis: More U.S. support for Israel is associated with more anti-American terrorism.

Main Data
We study the empirical relationship between anti-American terrorism and U.S. support for Israel in a time-series (Section 4) and a panel setting (Section 5). For both types of analysis we use the same measures of anti-American terrorism and U.S. support for Israel, which we describe below.

Anti-American Terrorism
Anti-American terrorism is measured by the number of terrorist attacks by citizens of another country against U.S. interests. Transnational anti-American terrorist attacks may be directed at diplomatic (e.g. embassies), military, commercial or non-official (e.g. tourists) targets. For instance, we consider transnational terrorism such as the 1988 bombing of a U.S. military recreational club in Naples (Italy) by the Japanese Red Army (where the attack is assigned to Japan). Terrorist attacks carried out by U.S. citizens within the U.S. that only victimize other U.S. citizens are, by contrast, not considered; a prominent example of this kind of domestic terrorism is the 1995 Oklahoma City bombing.
The data come from the International Terrorism: Attributes of Terrorist Events (ITERATE) dataset (Mickolus et al. 2016). This dataset defines (transnational) terrorism as the use (or threat of use) of anxiety-inducing, extra-normal violence for political purposes by any individual or group, whether acting for or in opposition to established governmental authority, when such action is intended to influence the attitudes and behavior of a target group wider than the immediate victims and when, through the nationality or foreign ties of its perpetrators, its location, the nature of its institutional or human victims, or the mechanics of its resolution, its ramifications transcend national boundaries (Mickolus et al. 2016). ITERATE excludes terrorist attacks against combatants (e.g. on U.S. troops that act as an occupying force in Iraq); however, it includes attacks against U.S. military forces that act as peacekeepers (e.g. during the 1983-1984 Lebanon intervention). Figure 1 displays the patterns of anti-American terrorism for the 1970-2014 period. As also discussed in LaFree, Yang, and Crenshaw (2009), there was more anti-American terrorist activity during the Cold War era  when the United States were prominently attacked by leftwing terrorist groups hailing from Latin America (e.g. the Peruvian Shining Path) and Western Europe (e.g. the Italian Red Brigades). Many of these groups lost popular and external support after the end of the Cold War, explaining the relative decline in anti-American terrorism after 1990. Instead, after 1990 anti-American transnational terrorism was more strongly associated with Islamist violence (e.g. by Al-Qaeda).

Measuring U.S. Support for Israel
For our analysis we consider two dimensions of U.S. support for Israel, military support in the form of military aid and diplomatic patronage at the United Nations.
Military support for Israel is measured by U.S. military foreign assistance provided to Israel (in constant dollars) as a share of total (global) U.S. military aid in a specific year. For instance, U.S. military assistance includes grants and loans given by the U.S. via the Foreign Military Financing Program that is used to help countries purchase defense equipment produced in the United States as well as acquiring defense services and military training. The data on military aid come from the United States Agency for International Development (USAID) (USAID 2015). 8 As shown in Figure 2, on average the U.S. provided approximately US$ 3 billion in military aid to Israel, where military aid was in general relatively steady but with noticeable peaks in the 1970s. As a share of total military aid, U.S. assistance to Israel was more volatile, ranging from two percent in the early 1970s to 60 percent in the late 1970s to approximately 25 percent in 2014.
To indicate diplomatic patronage of Israel by the United States we consider the similarities in voting between the U.S. and Israel at the United Nations General Assembly (UNGA) in a given year. UNGA voting data are drawn from an update of Voeten, Strezhnev, and Bailey (2009). The UNGA voting variable can lie between 0 for complete disagreement between both countries and +1 for complete agreement. Figure 3 compares the diplomatic alignment between the United States and Israel with the alignment between Israel and the rest of the world on average. Clearly, the alignment between the U.S. and Israel has been markedly stronger than the alignment between the rest of the world and Israel at least since the 1980s, which is indicative of a special diplomatic relationship between Israel and the United States. Consequently, our measure of diplomatic patronage is defined as the difference between the diplomatic alignment between the United  States and Israel and between Israel and the rest of the world on average. We interpret higher values of this variable to coincide with stronger diplomatic patronage by the United States. 9

Time-Series Approach
The determinants of anti-American terrorism may be analyzed at either the country-level or the level of the international system. In this section we employ a system-level time-series approach. Earlier examples that use such an approach to uncover the structural determinants of (anti-American) terrorism are O'Brien (1996), Volgy, Imwalle, and Corntassel (1997), Sobek and Braithwaite (2005), Lizardo (2006) and Krieger and Meierrieks (2015). Indeed, taking on a system-level perspective may be appropriate in the context of the nexus between U.S. support for Israel and the patterns of anti-American terrorism. U.S. support for Israel takes place within the international system; it is an observable policy variable that can be considered to be independent from country-level conditions in the country of origin of a potential anti-American terrorist. In other words, anti-American attacks motivated by U.S. support for Israel can occur anywhere in the world and be carried out by terrorist actors hailing from any country. 10 Consequently, for a test of our main hypothesis we compile timeseries (system-level) data for anti-American terrorism, U.S. support and additional controls for the 1970-2014 period. The summary statistics of these variables are presented in Table 1.

Empirical Model and Controls
Our time-series estimation equation looks as follows: Here, terror refers to the global number of anti-American terrorist incidents in a given year t, while support refers to the q-th indicator of U.S. support for Israel (share of U.S. military aid to Israel or diplomatic patronage at the UNGA). Besides a well-behaved error term (ε), the model also includes a vector of controls (X) discussed below. All independent variables are lagged by one year to reduce simultaneity concerns. Given the count-data nature of the dependent variable, which only assumes discrete, nonnegative values, we estimate Equation (1) using a generalized linear model for count data, where the count is expected to come from a negative binomial distribution that accounts for the overdispersion of the dependent variable (Cameron and Trivedi 2013). As time-series data is likely affected by serial correlation, we report Newey-West standard errors (Newey and West 1987). Finally, the use of time series data raises the issue of non-stationarity; using non-stationary data may give rise to spurious regression problems (Enders 2010). Applying the unit root test by Phillips and Perron (1988), we find that all variables except our measures of diplomatic patronage are stationary (Supplementary Table 1), so that issue such as spuriousness and cointegration do not arise. The diplomatic patronage variables are first-differenced to achieve stationarity. 11 With respect to the control variables (X), we only include a small number of controls to retain sufficient degrees of freedom, given that our number of observations is rather small (T= 44). First, we control for U.S. dominance in the international political system. Sobek and Braithwaite (2005) argue that when the U.S. is dominant, terrorism becomes the most cost-efficient means to change the status quo (e.g. via a regime or territorial change) that is upheld by the United States (Sobek and Braithwaite 2005). As in Sobek and Braithwaite (2005, 143-144), to operationalize U.S. dominance we identify the chief revisionist rival of the U.S. in a given year. The chief revisionist rival is the country with the highest CINC score (Composite Index of National Capabilities score) and the lowest agreement with the U.S. in terms of United Nations General Assembly (UNGA) in a given year. 12 The CINC score aggregates six individual components of national material capabilities (iron and steel production, energy consumption, total population, urban population, military personnel and military expenditures) into a single value per state-year to reflect a country's power, using updated data from Singer (1987). As in Sobek and Braithwaite (2005), the product of the UNGA voting variable drawn from Voeten, Strezhnev, and Bailey (2009) and the CINC score is our measure of U.S. dominance.
Second, we include a variable that measures American power relative to the power of its main rival (power balance). It is operationalized as the ratio of the U.S. CINC score to the chief rival's CINC score. A larger value of this variable corresponds to the U.S. being more powerful compared to its main rival. We remain agnostic about the effect of changes in the power balance variable on anti-American terrorism. For one, it is possible that increases in U.S. power deter terrorism as terrorist organizations may shy away from attacking a powerful enemy for fear of retaliation (Sobek and Braithwaite 2005, 141). For another, it may also be possible that American rivals are more likely to use terrorism (as a foreign policy tool) to undermine the advantage of a more powerful U.S., implying a positive relationship between U.S. power and anti-American terrorism (Volgy, Imwalle, and Corntassel 1997).
Third, during the Cold War the U.S. was repeatedly targeted by left-wing terrorist groups which were used by non-Western countries (e.g. Libya, Cuba) as proxies to challenge the U.S. (O'Brien 1996;Sobek and Braithwaite 2005). The end of the Cold War curtailed this support while undermining the ideological foundations of many left-wing terrorist groups (Shughart 2006), possibly contributing to a decline in anti-American terrorism after the end of the Cold War. We thus expect the Cold War era to positively predict anti-American terrorism. The Cold War era is indicated by a dummy variable that takes the value 1 for the 1970-1990 period and 0 afterwards.
As a robustness check, we amend our baseline model with two additional controls. For one, we control for incidences of Republican U.S. presidencies (indicated by a dummy variable that is equal to 1 when a Republican is president and 0 otherwise). Koch and Cranmer (2007) find that governments of the left (for the U.S., Democratic governments) are more likely to be targeted by terrorism than governments of the right (Republican governments). They argue that the more hawkish policies of right-wing governments deter terrorist activity. However, the same hawkish policies may also incite anti-American resentment. We thus remain agnostic about the effect of this variable on anti-American terrorism. For another, we control for the number of international crises the U.S. is involved in for a specific year, using data from the International Crisis Behavior Dataset updated from Brecher and Wilkenfeld (2000). We expect a greater involvement in crises to predict anti-American terrorism, e.g. as anti-American terrorism is used by crisis actors hostile to the United States.

Empirical Results
The time-series regression estimates are reported in Table 2. Considering our main variables of interest, there is no systematic evidence in favor of our main hypothesis; as reported in the appendix (Supplementary Table 2), this is also the case when both support variables enter the model at the same time. In other words, we find no evidence that more military-financial or diplomatic support by the United States for Israelregardless of how this support is operationalizedtranslates into more anti-American terrorism at the system-level. This interpretation does not only follow from a lack of statistical significance associated with the estimates for military-financial or diplomatic support but also the lack of economic substantiveness. For instance, we can transform the regression coefficient associated with military assistance in model (1) of Table 2 into an incidencerate ratio (IRR) of IRR = 0.998, implying that a one-percent increase in U.S. military aid to Israel results in a reduction in the global count of anti-American terrorist attacks by only 0.2 percent.
Considering the controls, the findings reported in Table 2 suggest that anti-American terrorism is associated with factors other than U.S. support for Israel. First, consistent with Sobek and Braithwaite (2005), higher levels of U.S. dominance lead to more anti-American terrorism. This may indicate that terrorism becomes a likelier means of challenging the status quo when U.S. dominance obstructs more conventional ways to change it (Sobek and Braithwaite 2005). Second, higher levels of U.S. power (relative to its main rival) are negatively related to anti-American terrorism. This finding may be indicative of a deterrence effect. 13 Third, anti-American terrorism was less likely after the end of the Cold War. A lack of material support from non-Western countries and the reduced ideological standing of left-wing terrorist groups after 1990 possibly accounts for this finding, as also argued by Shughart (2006). Fourth, we find that Republican presidencies are associated with more anti-American terrorism, potentially as these presidencies tended to coincide with more aggressive foreign policy behavior. Finally, being involved in international crises was also conducive to anti-American terrorism. For instance, protests against U.S. crisis behavior may have also taken on the form of anti-American terrorism (e.g. terrorist attacks against American interests in Western Europe during the Vietnam War).

Panel Analysis
The time-series analysis of Section 4 provides no evidence in favor of our main hypothesis. At the system-level, more U.S. support for Israel does not result in more anti-American terrorism. However, the system-level perspective has been criticized for not accounting for the role of country-specific conditions in explaining anti-American terrorism. For instance, Neumayer and Plümper (2011, 8) favor a country-specific (time-series cross-sectional) over a system-level (time-series) approach, arguing that only the former perspective sheds "[. . .] light on the causes of anti-American terrorism. This is because while such terrorism is widespread across the world, there are also clear and stark differences in its extent across terrorists' countries of origin." Indeed, for our panel of 150 countries for the period 1970 to 2014, 14 51 countries produced no anti-American terrorism and 49 countries produced less than five anti-American terrorist attacks during the observation period. Thus, it may be meaningful to also consider the nexus between U.S. support for Israel and anti-American terrorism in a framework that explicitly accounts for country-specific factors.

Empirical Model and Controls
Our time-series cross-sectional estimation equation looks as follows: Here, terror refers to the number of anti-American terrorist incidents produced in country i in year t, while supportas in Equation (1) refers to the q-th indicator of U.S. support for Israel. These indicators are the same for all countries in the sample for a specific year, accounting for the fact that U.S. support for Israel is an observable policy variable that is independent from country-level conditions in the country of origin of a potential anti-American terrorist. Model (2) also includes a well-behaved error term (ε) and a vector of controls (X) discussed below. All explanatory variables are lagged by one year to reduce endogeneity concerns. We estimate model (2) running a series of negative binomial regressions (Cameron and Trivedi 2013). As above, this method is appropriate because our dependent variable is a count variable (the number of anti-American terrorist incidents per country-year) that only assumes discrete, non-negative values. Given the presence of overdispersion (i.e. the variance of the dependent variable is larger than its mean), a negative binomial model is preferred over a Poisson model (Cameron and Trivedi 2013).
Considering the vector of controls (X), we follow the literature on the determinants of terrorism (for a review, see Krieger and Meierrieks 2011) as well as on the determinants of anti-American terrorism in particular (e.g. Neumayer and Plümper 2011;Gries, Redlin, and Meierrieks 2015;Krieger and Meierrieks 2015). First, we control for the effect of economic development which is measured by (logged) real per capita income. The data come from the Penn World Table (Feenstra, Inklaar, and Timmer 2015). Richer countries may be less likely to produce anti-American terrorism due to the high opportunity costs of participating in terrorism (Krieger and Meierrieks 2011). We thus expect a negative correlation between a country's level of economic development and the level of anti-American terrorism .
Second, we control for (logged) population size. The data are again drawn from the Penn World Table. Countries with larger populations are expected to produce more anti-American terrorism because they exhibit a larger pool of (potential) terrorists, terrorist supporters, and (U.S.) victims. Higher policing costs in larger countries may also favor the emergence of terrorism. Indeed, the empirical consensus on the determinants of terrorism is that population size ought to correlate positively with terrorism (Krieger and Meierrieks 2011).
Third, we consider the role of military aid by the United States as a potential determinant of anti-American terrorism. It is measured by the amount of U.S. military aid (in constant U.S. dollars) provided to a specific country divided by local population size to account for scale effects; the data on military aid are drawn from USAID (2015). As already argued above, in discussing the strategic logic channel, Neumayer and Plümper (2011) find that countries receiving more military aid are also more likely to produce anti-American terrorism. Generalizing the ideas underlying the strategic logic channel (i.e. abstracting from the case of Israel), we can expect military aid by the United States to strengthen any local government opposed by any local terrorist group, which creates a strategic incentive for the local terrorist group to attack the U.S. so as to weaken U.S. support for the local government (Neumayer and Plümper 2011). We thus expect military aid by the United States to be positively related to anti-American terrorism.
Fourth, a country's regime type may also matter. This variable is measured by the Polity2 score from the Polity IV Dataset of Marshall, Gurr, and Jaggers (2015), which ranges from −10 (strongly autocratic regime) to +10 (strongly democratic regime). For one, more democratic countries may see less (anti-American) terrorism because democracies offer (cost-efficient) nonviolent means to achieve political change, making terrorism a less attractive option (e.g. Krieger and Meierrieks 2011). For another, democracies are also less likely to counter terrorism effectively due to constraints on the executive (e.g. respect for civil liberties), which may make democracies likelier breeding grounds of terrorism (e.g. Krieger and Meierrieks 2011). Thus, we remain agnostic about the expected effect of democracy on anti-American terrorism.
Fifth, we control for state failure, indicated by a composite state failure index capturing the intensity of revolutionary and ethnic wars, adverse regime changes, genocides and politicides published by the Political Instability Failure Task Force (PITF Global Report 2015). We expect state failure to positively predict anti-American terrorism. For instance, failing states exhibit only weak law enforcement and military capabilities, making it easier for terrorist groups to operate (Piazza 2008).
Sixth, we again include a control for the Cold War era (indicated by a dummy variable that takes the value 1 for the 1968-1990 period and 0 thereafter). As with the time-series analysis of Section 4, we expect the Cold War era to positively predict anti-American terrorism.
Finally, as robustness checks we amend our baseline model with two additional controls accounting for geographical distance to the United States 15 and the level of trade openness (Penn World Table data). We expect the former variable to negatively predict anti-American terrorism (as distance ought to increase the costs of carrying out this type of terrorism) and the latter to positively predict it (e.g. as higher levels of openness coincide with stronger antiglobalization resentment that results in anti-American terrorism; for a further discussion see Krieger and Meierrieks 2015).

Main Results
Our panel analysis results are reported in Table 3. Considering our main variables of interest, we findconsistent with the time-series evidence presented aboveno systematic evidence that more U.S. support for Israel leads to more anti-American terrorism. This finding holds regardless of how U.S. support for Israel is operationalized; as shown in the appendix (Supplementary Table 3), it furthermore holds when both support variables enter the model at the same time. That is, there is again no support for our main hypothesis.
The results for the control variables are remarkably similar across the various model specifications. These results provide insights into which economic, demographic and political variablesrather than U.S. support for Israelare indeed relevant to explaining anti-American terrorism. First, we do not find that richer countries are more likely to produce anti-American terrorism. This is consistent with the empirical mainstream that economic underdevelopment tends to be a weak predictor of terrorism (Krieger and Meierrieks 2011). Second, more populous countries produce more anti-American terrorism. This finding is also consistent with the empirical mainstream on the determinants of terrorism (Krieger and Meierrieks 2011) in that more populous countries have larger pools of potential terrorists and victims so that they ought to see more anti-American terrorism due to scale effects. Third, as previously found by Neumayer and Plümper (2011) and Gries, Redlin, and Meierrieks (2015), more military aid by the United States leads to more anti-American terrorism originating in aid-receiving countries. 16 Potentially, this is because local terrorist groups use anti-American terrorism to curtail U.S. support for local governments (Neumayer and Plümper 2011). Importantly, this result suggests that the strategic logic channel matters at the (local) country-level rather than the system-level. Fourth, regime type does not matter to anti-American terrorism, meaning that neither autocratic nor democratic countries are more likely to generate anti-American terrorism. Fifth, anti-American terrorism is more likely to originate from failing states, which is consistent with Piazza (2008) who argues that these types of states offer terrorist groups favorable conditions (e.g. weak and corrupt law enforcement) under which to operate. Sixth, as expected, anti-American terrorism was more likely during the Cold War (e.g. due to support for it by non-Western countries). Finally, while a country's level of trade openness does not predict anti-American terrorism, terrorism against the United States is less likely to originate from countries that are further away from the U.S. in terms of geographical distance, potentially because distance increases the costs of carrying out attacks and reduces the availability of American targets.

Further Panel Analyses
Our various measures of U.S. support for Israel are not country-specific butas variables measured at the system-levelidentical across countries. Still, we can argue that these support variables should not uniformly affect anti-American terrorism for all countries in our sample. For instance, in countries such as Egypt (due to its proximity to Israel) U.S. support for Israel may be a much  stronger catalyst of anti-American terrorism than in, e.g. Colombia. In other words, there may be spatial (or country-specific) heterogeneity associated with terrorist responses to U.S. support for Israel. In the following, we consider two factors that may explain why U.S. support for Israel incites anti-American terrorism more strongly in specific countries.
1. We argue that being from a MENA country (i.e. from a country located in the Middle East or Northern Africa) may make terrorists particularly responsive to (changes in) U.S. support for Israel. For one, these countries are geographically close to Israel. For instance, this means that many Arab-Palestinian operate and recruit in these countries (e.g. the Popular Front for the Liberation of Palestine -General Command in Syria or the Palestinian Liberation Organization in Lebanon and Tunisia). For another, many MENA countries are stakeholders in the Arab-Israeli conflict which is characterized by ethnic-ideological tensions (Zionism vs. Arab nationalism) and international warfare (e.g. the Six-Day War of 1967). Citizens of MENA countries may consequently more strongly resort to anti-American terrorism due to the (perceived) one-sidedness of the United States in the Arab-Israeli conflict. 2. As already argued in this paper, there is a distinct religious dimension to the Israeli-Palestinian/Arab conflict in the form of a conflict between Judaism (supported by the U.S. as a predominantly Christian country) and Islam. On the part of Israel's enemies, this conflict can consequently be framed as a 'holy war' to dispel 'infidels' from Islamic holy sites and to stop the 'oppression' of Muslim by non-Muslims. Thus, the responsiveness to U.S. support for Israel may increase with a country's Muslim population share.
To estimate whether these two factors (location in the MENA region or the Muslim population share) matter, we estimate two variants of model (2) that are specified as follows: In models (3a) and (3b) we are primarily interested in whether the regression coefficients associated with the various support variables sufficiently change when we compare (i) whether a country of origin of a terrorist committing anti-American terrorism is or is not located in the MENA region or (ii) the role of a country's Muslim population share. 17 We formally examine this by running a series of tests of coefficient equality, i.e. tests with the null hypothesis that β 11 = β 21 for model (3a) and β 31 = β 41 = β 51 for model (3b). We report our empirical findings for models (3a) and (3b) in Table 4. In short, the results suggest that the size of the Muslim population share does not matter. Countries with a larger Muslim population share do not produce more anti-American terrorism given a specific level of U.S. support for Israel. Rather, regardless of how large the Muslim population share is, there is never a statistically significant association between the support variables and the emergence of anti-American terrorism.
However, there is a difference between MENA and non-MENA countries. More U.S. military aid and U.S. diplomatic patronage of Israel result in more anti-American terrorism originating from MENA countries but do not affect anti-American terrorism originating from non-MENA countries. To study the substantiveness of the associated effects, we can furthermore calculate the associated IRR. For instance, for model (1) of Table 4 the calculated IRR suggest that a tenpercent increase in U.S. military aid leads to (ceteris paribus) a 32 percent increase in the number of anti-American terrorist attacks originating in MENA countries; the same increase in aid, by contrast, only results in a 7 percent rise (that is not statistically significant) in the number of anti-American terrorist attacks from non-MENA countries. In sum, this latter finding may then suggest that the geographical proximity to Israel as well as ethnic, religious and political circumstances that are specific to the MENA region make citizens from MENA countries more responsive to U.S. policies regarding Israel and consequently more likely to engage in anti-American terrorism.

Conclusion
Over the last decades, terrorist organizations with various ideological profiles have repeatedly justified attacks against U.S. interests as punishment for the (purported) special relationship between the United States and Israel. What is more, the idea of a nexus between the United States' support for the state of Israel and its vulnerability to transnational terrorism has also been controversially discussed in the scholarly and public arena, especially in the context of the 9/11 attacks. Motivated by these considerations, for the 1970-2014 period we study whether U.S. support for Israel indeed induces anti-American terrorism at the level of the international political system (using time-series data) and at the country-level (using time-series cross-sectional data). We argue that the link from the U.S. support for Israel to anti-American terrorism may run via the (i) strategic logic channel, where terrorist groups may resort to anti-American terrorism to make U.S. support for Israel more costly, ultimately shifting the balance of power in the terrorists' favor or (ii) the support channel, where terrorist organizations may punish the United States for its (perceived) one-sided support for Israel to garner support from anti-Semitic or anti-Imperialist segments of society (popular support) as well as state and terrorist actors that are hostile to the United States and Israel.
Our empirical results indicate that regardless of which level of analysis or which dimension (military or diplomatic) of U.S. support for Israel we consider, more U.S. support does not lead to more anti-American terrorism. Instead, a variety of other economic, political and demographic factors are found to sway anti-American terrorism at both levels of analysis. For instance, while U.S. military aid to the Israeli government does not affect anti-American terrorism, U.S. military aid to local governments fuels it, suggesting that circumstances in the terrorists' country of origin influence their calculus more strongly than (potentially more abstract) international conditions.
Our empirical study was aimed at identifying a systematic effect of U.S. military support for and diplomatic patronage of Israel on anti-American terrorism. We were not able to detect such an effect. As an important qualification, however, we did find some (preliminary) evidence that for the MENA region a favorable U.S. policy stance towards Israel may have indeed contributed to anti-American terrorism. For instance, this finding is relevant in the context of the 9/11 attacks, where all perpetrators hailed from MENA countries. However, more research is necessary to understand the underlying mechanisms that appear to be specific to the MENA region more clearly. Notes 1. Transnational terrorism means that more than one country is affected by a terrorist incident. For instance, the 9/11 attacks were transnational because foreign terrorists attacked on U.S. soil. 2. Given its delicate and somewhat polemic nature, the study by Mearsheimer and Walt (2006) has drawn a fair share of criticism, e.g. by Lieberman (2009). 3. For instance, this three-way strategic interaction between a local government, its external sponsor and a terrorist group hostile to both the local government and its sponsor is modelled by Addison and Murshed (2005) and Neumayer and Plümper (2011). 4. For instance, this internationalization already becomes apparent in the fact that Khalid Sheikh Mohammed is a Pakistani citizen. 5. Pape (2003) provides an example where such a coercive strategy was successful. In 1983 Hezbollah began a (suicide) terrorist campaign against a U.S.-led multinational peacekeeping mission during the Lebanese civil war. The campaign ended with the withdrawal of U.S. forces due to the political pressure the campaign had created in the United States (Pape 2003). 6. Anti-Semitism as a motive for anti-American terrorism is discussed in Sedgwick (2004). For an example of anti-Semitic thinking associated with radical Islamist groups, take the following quote from Bin Laden's (2002) 'Letter to America':'[Y]ou [the U.S.] build your economy and investments on usury. As a result of this, in all its different forms and guises, the Jews have taken control of your economy, through which they have then taken control of your media, and now control all aspects of your life making you their servants and achieving their aims at your expense [. . .].' 7. Information on state sponsorship of terrorism is drawn from various issues of the Country Reports on Terrorism published by the U.S. State Department. 8. As a robustness check, we also use the (logged) amount of U.S. military foreign assistance provided to Israel in a given year as an alternative indicator of U.S. military support for Israel. As reported in the appendix (Supplementary Tables 2 and 3), using this alternative indicator does not change the empirical findings reported in the main text. 9. As another robustness check, we also use the level of diplomatic alignment between the United States and Israel at the UNGA as an alternative measure of the special diplomatic relationship between Israel and the United States. As reported in the appendix (Supplementary Tables 2 and 3), employing this alternative indicator of diplomatic patronage does not affect the findings reported in the main text. 10. Take Al Qaeda as an example, a group clearly motivated in part by American support for Israel. The group attacked American targets in the 1990s and 2000s in various parts of the world (e.g. the 1998 attacks on the U.S. embassies in Kenya and Tanzania, the 2000 attack on the USS Cole in Yemen). Its operatives hail from numerous countries in the Middle East (e.g. Saudi Arabia, the United Arab Emirates and Egypt) but also non-Muslim majority countries (e.g. Canada, the United Kingdom, India and Kenya). 11. Using the diplomatic patronage variables in levels (i.e. without taking the first-difference), however, does not yield results that are different from the ones reported below (results available upon request).