![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
Abstract
Analysing the relationship between the provision of military support and the probability of becoming the target of a terrorist attack, this paper contributes to the literature on the causes of transnational terrorism. We find that deployment of military troops of country X in country Y increases the probability of a terrorist attack on citizens of country X by a terrorist group located in country Y. Exporting weapons to country Y seems to increase the probability of being attacked by the terrorists of this country Y as well. Deploying materials, however, does not seem to significantly influence the probability of attack. Including lagged values for our military support variables ensures that the causality direction is from military support to terrorist attacks. Moreover, these results indicate that while the effect of military deployment on the probability of attack lasts for more than 1 year, the effect is rather short-lived.
Acknowledgement
The insightful and constructive comments by an anonymous referee are gratefully acknowledged. The usual disclaimer applies.
Notes
1 In this paper, we will not study domestic terrorism but rather focus on transnational terrorist attacks that are defined by Enders and Sandler (Citation2006, 7) as ‘incidents in one country involving perpetrators, victims, institutions, governments, or citizens of another country’.
2 The large number of victims in the attacks of 9/11 indicates that media attention and hence publicity cannot be considered the only goal of terrorists (Hoffman Citation2002).
3 Our measure of military support differs from the notion of ‘military aid’ as defined by the US Greenbook as loans or grants, military transfers or training of the military personnel of the recipient country (Bapat Citation2011).
4 The variable equals 0 if the 2 countries share a border.
5 Domestic terrorism was eliminated after having determined the nationality of the perpetrator or the organisation that claimed the attack. Attacks where the perpetrator and the victim have the same identity are classified as domestic and are thus dropped, as our focus is on transnational terrorism only. When the nationality of the perpetrator and the victim is different, the observation is classified as transnational terrorism and kept in the data-set. The location of the attack is not taken into account. Attacks where the perpetrator is unknown are dropped from the data-set.
6 All quantitative analyses were carried out using the software STATA (Statacorp Citation2009).
7 Original coefficients can be found in Table A1 of Appendix 1. Note that the coefficients of a negative binomial panel estimate cannot be interpreted in the same way as the coefficients that one obtained from an OLS.
8 In the remainder of the paper, we will work with the dependent variable measuring the number of attacks. Results for casualties are similar, taking into account that the level of significance of the independent variable measuring deployment then increases from the 5% to the 1% level and that the dummy variable reflecting weapon export loses its significance.
9 Shehzad Tanweer, one of the terrorists of the London bombings in a 2005 video message.
10 The correlations between deployment and its first lag and between the two lags are particularly worrisome, with values of 0.9 and 0.91, respectively.
11 Koyck (Citation1954) The Koyck transformation is quite an old method but is still very influential and frequently used, as explained in Franses (Citation2004).
A good overview of the technique can also be found in Gujarati (Citation2003), 665–669.
12 As a robustness check, we also test the null hypothesis that lambda is equal to 1 to make sure that a longer effect can be excluded. This null hypothesis is firmly rejected as the calculated value is −15 and thus falls outside the area where the null hypothesis can be accepted (−2; 2). Additionally, including all control variables in the Koyck transformation does not change the results. The coefficient lambda of both Koyck transformations even decreases below two.