![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
Scholars maintain that, similar to insurgency, terrorist violence is precipitated by both relative deprivation and state weakness. Yet aggrieved minority groups within a country should turn to terrorism when they are weak relative to the state rather than strong. Empirical evidence shows minority group discrimination and fragile political institutions to independently increase domestic terror attacks. But it remains unclear whether grievances drive domestic terrorism in both strong and weak states. Using data from 172 countries between 1998 and 2007, we find that for strong states the presence of minority discrimination leads to increased domestic terrorism, while for weak states the presence of minority discrimination actually leads to less domestic terrorism. Consequently, increasing state capacity may not be a panacea for antistate violence, as nonstate actors may simply change their strategy from insurgency or guerrilla warfare to terrorism. Efforts to reduce terrorist violence must focus on reducing grievance by eliminating discriminatory policies at the same time that measures to improve state capacity are enacted.
Notes
1 Sudan, Central African Republic, and Chad experienced an average of 0.63 attacks per year between 1970 and 1990. For the same period, Israel, Spain, and the United Kingdom suffered from an average of 42.63 yearly attacks of domestic terrorism.
2 It is important to note that minority discrimination is only one of the many drivers of terrorism. Throughout history, desperate individuals have resorted to terrorism as a strategy for achieving their political objectives. Several types of grievances have driven rebels. For example, physical repression by dictators, income inequality, desire for self-rule, and fear of losing social privilege are some of the grievances that have driven jihadist, revolutionary, separatist, and right-wing terrorism. Even the perception of discrimination and social alienation might constitute grievances. This study uses a state’s discriminatory policies as an important form of grievance, while state capacity gets at opportunity.
3 We focus on domestic terrorism because it is a far more frequent occurrence than international terrorism, although the latter may generate more media attention (Abadie Citation2006). In fact, domestic terrorism represents by far the greatest type of terrorist violence observed (see ). Of the nearly 60,000 terrorism incidents recorded in the Global Terrorism Database between 1970 and 2007, local, within-country groups, produce 78% of the attacks.
4 lso see Wimmer, Cederman, and Min (Citation2009), who note that rebellion is more likely in states that exclude large portions of the population on the basis of ethnic background.
5 These are regimes characterized by some measure of political and civil liberties but concomitantly by a deficient rule of law and widespread human rights violations (Feldmann and Perala Citation2004). This finding supports the theoretical expectation that terror is more likely to be adopted as an opposition strategy in political settings where resource mobilization is possible but where peaceful protest generally produces few tangible results.
6 Lai (Citation2007) finds that transnational terrorism originates in weak states, and most states sending out transnational terrorists also experience civil wars. Therefore, his logic of weak states as “swamps” for transnational terrorism is empirically supported. Piazza’s (Citation2008) study also concentrates on transnational terrorism and not its domestic counterpart. It remains possible that weak states offer space for the emergence of transnational terrorist groups, and strong states propel the growth of domestic terror cells when significant grievances exist.
7 We choose the time period of 1998–2007 for specific reasons. Pape, Ruby, Bauer, and Jenkins (Citation2014) have identified several inconsistencies in the Global Terrorism Database (GTD) collection method and recommended breaking the GTD data into four distinct data sets, each based on a consistent collection method. Those are: 1990–1997, 1998–2007/2008, 2008–2011, and 2011–2014. The Enders et al. (Citation2011) data on domestic terrorism used in our study are based on GTD and are available for the period of 1970 to 2007. We decided to test our hypothesis on data for the years between 1998 and 2007 because, firstly, data on many of our explanatory variables are available for this time period. World Bank data on Government Effectiveness and Lack of Corruption are available from 1996 onwards. The State Fragility Index data start from 1996. Secondly, GTD provides limited data on terrorist incidents for the year 1993. The results of our empirical models might be biased if we include 1993 in our analysis.
8 Access to the raw GTD database, along with descriptions of count methods and operationalization of terrorism, is available at http://www.start.umd.edu/gtd/.
9 Terrorism is the premeditated use or threat to use violence by individuals or subnational groups against noncombatants in order to obtain a political or social objective through the intimidation of a large audience beyond that of the immediate victims.
10 Uncertain observations include incidents involving insurgency or guerilla warfare, internecine conflict, mass murder, and criminal acts.
11 We include civil war defined as 1,000 battle deaths as a dummy variable in all the models. Some groups engaged in civil war also use terrorism as a strategy. Controlling for civil war increases our confidence that large-scale political conflict within countries isn’t driving our results. However, we ran models excluding the civil war dummy, and our results remain unchanged.
12 Access to the raw Uppsala/PRIO database, along with descriptions and operationalizations of civil war and interstate war, is available at http://www.prio.no/Data/Armed-Conflict/.
13 A few studies (De la Calle and Sánchez-Cuenca 2012; Enders, Hoover, and Sandler Forthcoming; Freytag, Krüger, Meierrieks, and Schneider Citation2011; Lai Citation2007) find an inverted U-shape relationship between per capita and terrorism, which could suggest that domestic terrorism remains low in weak states, increases in moderately strong states, but returns to low levels in very strong states. In fact, Fearon and Laitin (Citation2003) use GDP as a measure of state strength. Our concern is that GDP more directly proxies for economic opportunity and not state strength. The inverted U finding using GDP may reasonably indicate a relationship about opportunity and relative deprivation and not necessarily state strength. We believe our measures more directly measure state strength and its effect on domestic terrorism.
14 The CShapes data set was used to create the inverted distance interdependence matrix (Weidmann, Kuse, and Gleditsch Citation2010).
15 The dependent variable has a mean of 5.29 and standard deviation of 26.37.
16 As the data on our dependent variable consists of 69% zeroes, a zero-inflated model could be appropriate. We ran a series of Zero Inflated Negative Binomial (ZINB) models, and the results generally support our theoretical expectations. Still, we remain skeptical of the zero-inflated model for several reasons. First, one has to assume with the ZINB model that some observations in our data set (so some countries during some years) have a zero probability of experiencing domestic terrorism. We are hesitant to make such an assumption because almost every country suffers terrorism at some point in history. Drakos and Gofas (Citation2006), in their piece on underreporting bias in quantitative studies of terrorism, argue against full specification of the inflated equation in zero-inflated negative binomial modeling and recommend instead including only covariates associated with “certain-zero” countries: regime type. They assume that certain-zero countries appear to be so in the data because they lack free media that would report on terrorist events. However, the Global Terrorism Database (GTD) data collection method is robust to this type of bias, since it does not solely depend on local media. In the absence of a strong theoretical justification for modeling the zero observation, we are not confident of using ZINB models. However, our findings are robust to different types of specifications (ZINB, Negative Binomial, Random Effect NB, and Fixed Effect NB). This strengthens our confidence that our empirical results are not strongly influenced by our estimator choice.
17 Interestingly, separate analyses not included here show the square of the percentage of discriminated population to have a statistically significant and negative impact on domestic terrorism. Domestic terrorism increases as more people are politically discriminated against, but domestic terrorism decreases as the percentage of discriminated-against population becomes very high. This finding does support our theoretical expectation that terrorism is a strategy of the weak. As the number of people facing state-led discrimination increases, rebel groups representing such populations would likely emerge to engage the state more directly and conventionally. We intend future analyses to examine more carefully when minority groups transition from terrorist violence to insurgency.
18 A control for the US military interventions in Afghanistan and Iraq could not be included in models using the State Fragility Index (Reversed) since the index does not record values for either state during these years.
19 We think it likely that countries with low discrimination drive the coefficient value for state capacity in the unconditional models in and .
20 We ran Wald tests on each of our interaction models to assess whether restricting parameters on the interaction term to 0 significantly harms the fit of our models. In every case, the Wald tests indicate that the models with the interaction terms improve the overall fit. As an additional test, we also include quasi-likelihood information criteria (QIC) scores for each model in –. As you can see, for most models the QIC scores are substantially lower for the interactive models.
21 As many groups engaged in civil wars use terrorism as strategy, some would argue that we are including violence practiced by such groups in our models. Therefore, we deleted all the incidents of terrorism for countries engaged in civil war (1,000 or more deaths), substituted those country-years with zero, and ran all our models. The results remained unchanged. Moreover, Enders et al. (Citation2011) have removed possible incidents relating to civil war from the data set (see Footnote 8).
22 We provide plots for only two of our interaction-term models due to limited space. These figures in the appendix are meant to supplement our marginal effect figures ( and ).
23 There are potential problems with the SFI and World Bank measures. The State Fragility Index could be endogenous to conflict (Coggins Citation2015), and the World Bank data on Government Effectiveness and Lack of Corruption are based on surveys. Relatively Extractive Capacity, on the other hand, takes into account a state’s tax-collection capacity more objectively than the other measures.