Abstract
A prominent theory holds that groups may use terrorism in order to provoke governments into undertaking repression that alienates the population. However, virtually no studies have addressed the central puzzle of this provocation logic: why states would actually fall into this trap, if doing so can backfire. This study seeks to address this puzzle by suggesting conditions under which states would respond to terrorism with repression. I argue that states with limited bureaucratic capacity are more prone to using repression after terrorist incidents, as their ability to selectively crack down is inhibited by their more limited capability for controlling, monitoring, and collecting revenue from their populations and for collecting intelligence on suspected terrorists. Using a cross-national analysis with data from 1981 to 2011, I find it is low-capacity states which are most likely to respond to terrorism with repression, while constraints on executive authority have no clear effect.
Although scholarly work has shed much light on the causes and consequences of terrorism, much less effort has been made to assess how and when governments respond to incidents of terror. This is a critical lacuna if we are to understand the effectiveness of terrorism and of counterterrorism; knowing how states react, and under what conditions, will allow us to better address the question of what they might do differently. The matter is of great theoretical significance as well, since many explanations of why rebel groups choose terrorism have made assumptions about how governments will respond, despite the relative scarcity of empirical evidence on government retaliation (Lake 2002; Kydd and Walter 2006). 1
In particular, much of the rationalist literature on terrorism posits that groups may choose to victimize civilians as a means of provocation, in order to bait the government into taking disproportionately repressive actions that convince the populace that it “is evil and untrustworthy and must be vigorously resisted”—or, at the very least, that it is not worth supporting (Kydd and Walter 2006, 69). The basic logic of provocation theory holds that terrorists can use the strength of the government against it by goading it into adopting harsh countermeasures that harm civilians, or at the very least impose upon their liberties. Repression can push extremists into action by motivating them to seek revenge; however, perhaps the most important consequence, scholars have argued, is that it radicalizes moderates by revealing the true, brutal nature of the regime in power (Lake 2002; Kydd and Walter 2006; Bueno de Mesquita and Dixon 2007). Holmes (2001) argues that government repression aimed at counterterrorism alienates the population and ultimately runs the risk of undermining the state’s legitimacy. Empirically, Walsh and Piazza (2010) show that, indeed, repression often fuels rather than subdues terrorism.
It is puzzling, then, why governments would choose to respond repressively to instances of terrorism, as provocation theory suggests they will, if doing so can be counterproductive. Not only does the use of repression carry the potential for alienating the population, as Bueno de Mesquita (2005) and Siqueira and Sandler (2006) argue, repression can also benefit terrorist groups by harming the state’s economy and thus making terrorism relatively less unattractive for individuals than would be the case in better economic conditions. More generally, repression also runs the risk of sparking an international backlash, which may even render the country a less attractive trading partner and destination for foreign direct investment (Hafner-Burton 2005; Franklin 2008).
It is the goal of this study to address this puzzle, as theorizing about exactly when governments will actually be provoked has been limited. In particular, I argue that governments’ bureaucratic–administrative capacity influences their propensity to use repression in response to terrorism by shaping their options for counterterrorism. Low-capacity states have less capability for extracting resources and redressing grievances, controlling and monitoring their populations, and collecting intelligence on suspected terrorists, which constrains their ability to selectively crack down and thus encourages them to use less discriminating counterterror tactics (Kalyvas 2006).
Ultimately, I find that states do frequently respond to terrorism with repression, and that this propensity is, consistent with previous research, related to the frequency of terrorist incidents. Additionally, I find that this relationship is a conditional one. States with low bureaucratic capacity are more likely to be provoked by terrorism, while high-capacity states are less likely to be. By contrast, constraints on executive power are shown to have very little effect on the use of repression in response to terrorism. Finally, when I separate domestic and transnational terrorism, I find that bureaucratic capacity remains the most important moderator of the relationship between terrorist incidents and repression, particularly for transnational terrorism. The finding that state capacity is a much stronger determinant of the use of repression after terrorist incidents than are constraints on the executive supports the provocation thesis, as it suggests that even states with few limitations on their use of coercive violence are unlikely to “take the bait” unless their ability to use more selective means of coercion is also limited.
This article’s contributions are twofold. First is its evaluation of the logic of provocation theory. If the results do not show that states actually do respond forcefully to terrorist incidents, then it may be necessary to reconsider provocation as a rational strategy of terrorism (Kydd and Walter 2006). At the same time, however, it is puzzling why governments would take the bait—why they would be complicit in the provocation. I provide leverage on this question by suggesting circumstances in which governments may forcefully strike back, despite the risk, due to their limited ability to conduct more discriminate responses.
Second, studying how and when states respond to terrorism provides insights into why groups choose terrorism to begin with. For example, scholarly work arguing that democratic rule promotes terrorism makes the assumption that democratic governments face institutional and public opinion constraints that limit how they can react (Chenoweth 2013), while the literature on state weakness and terrorism emphasizes that poor and failed states face the problem of limited capability that restricts their freedom of action (Coggins 2015), with both arguments coming to the conclusion that states which are limited in their options for responding are vulnerable targets for terrorists. However, these assumptions are not explicitly tested and thus deserve scrutiny if we are to understand the use of terrorism.
I begin with an overview of what the literature has suggested—or implicitly assumed—regarding why states might respond to terrorism with repression, as well as past findings on the relationship between terror and repression. In light of the potentially counterproductive effects of repression, I then suggest conditions under which states might still use repression, despite the risks. In particular, I emphasize the role of state capacity, which I argue helps address the central puzzle of the provocation theory. I then proceed with a discussion of the data used in this study, followed by a quantitative analysis of these data. Finally, I consider the implications of my findings for scholars’ understanding of state responses to terror and suggest future avenues through which the provocation logic of terrorism might be researched.
Literature Review: Terrorism and Repression
Scholarly studies that have systematically sought to establish how states respond to terrorism are few in number. Piazza and Walsh (2009), Robison (2009), and Dreher, Gassebner, and Siemers (2010) have found that transnational terrorism often results in the use of state terror, while Shor et al. (2014) present evidence suggesting that domestic terrorism is also related to the restriction of physical integrity rights. On the use of torture in particular, Conrad et al. (2016) find that military agencies are more likely than civilian agencies to use torture in response to transnational terrorism and that this relationship is strongest in democracies. However, while these suggest that governments can be goaded into a brutal response, with the exception of Conrad et al. (2016), these studies do not consider whether terrorism’s effect on repression is conditional upon other variables. Moreover, they do not address the central puzzle posed by the provocation logic of terrorism.
Why would states respond to terrorism with repression? Scholars writing on the effectiveness of terrorism have posited that the response of the government will be related to the severity of the terrorist threat. In effect, studies have hypothesized that terrorists can be too successful—that is, their success at the tactical level of being able to carry out attacks can lead governments to conclude that the rebels are not trustworthy bargaining partners (Fortna 2015). Broadly, a number of scholars have argued that concessions or a failure to respond to destructive incidents of terrorism might be seen by governments as a signal of weakness or a lack of resolve (Lapan and Sandler 1988; Bueno de Mesquita and Dixon 2007). Wood and Kathman (2014) similarly suggest that an escalation of terrorist attacks may have the paradoxical effect of signaling to the government that the rebels are actually too strong, and that since the balance of power is shifting in their favor, no credible long-term deal can be struck with them until they are beaten back. Abrahms (2006) comes to a similar conclusion via a psychological rather than a strictly rationalist mechanism and posits that terrorism is an ineffective means of achieving strategic goals because the act of repeatedly killing civilians and destroying property leads targets to conclude—rightly or wrongly—that the aims of the rebels are insatiable and that they cannot be bargained with but instead must be defeated.
Thus, an especially potent terrorist threat can be expected to lead to a more vigorous response first of all by sending too strong a signal of strength, which can create the perception that although the rebels’ threats are credible, their promises to cease violence and uphold any deal that is struck are not; and second, by creating a perception that the rebels are implacable enemies who cannot be appeased with concessions. Empirically, this suggests that more incidents of and greater casualties from terrorism will invite a more forceful response, as it is more likely that governments facing more destructive and more frequent attacks will feel compelled to subdue the threat, despite the risks of overreaction. Additionally, governments are likely to see exceptionally active and lethal terrorist organizations as implacable and extreme and thus to abandon hopes of negotiation until the balance of power shifts further in the state’s direction (Abrahms 2006; Wood and Kathman 2014). Indeed, de Figueirado and Weingast (2001, 6) suggest that governments “will only [repress] if the threat of future violence is credible and the prospect of stemming that violence is sufficiently high.” This implies that, all else being equal, high levels of terrorist incidents and casualties inflicted by terrorism increase the probability of repression, as previous studies have found (Piazza and Walsh 2009; Robison 2009; Dreher, Gassebner, and Siemers 2010; Shor et al. 2014).
Nevertheless, this provides at best a partial explanation for why governments would be provoked. While policy makers in states facing a particularly severe threat from terrorism may be compelled toward harshness in order to crack down against what is perceived to be a dangerous and implacable foe, this still does not address the central puzzle, as it does not explain why governments could not use more targeted means. Thus, I posit that states with low bureaucratic capacity, because of their more limited means of information collection and sophisticated policy implementation, will be the most susceptible to provocation.
State Capacity: Administrative Reach, Information, and Provocation
According to one line of argument, terrorism is a “weapon of the weak,” and we should expect to see terrorism occurring most frequently in strong states, where open insurgency is ruled out because of target hardening (Crenshaw 1981; Pape 2003). On the other hand, however, states with weak institutions may be more prone to terrorism simply because they are less able to find and crack down on the terrorists and respond to popular grievances (Goodwin 2001; Collier and Hoeffler 2004).
To reconcile this, one can distinguish “bureaucratic/administrative” capacity—the extent to which states are able to keep tabs on the population and coerce it via nonmilitary means—from “military” capacity, which refers simply to the strength of the armed forces (Hendrix 2010; Hendrix and Young 2014). States with high levels of bureaucratic–administrative capacity are characterized by the presence of competent, technocratic bureaucratic structures in which career advancement is based upon merit, and which have significant institutional presence across the country’s territory that allow the state to exert its authority (Weber [1919] 1946; Mann 1993; Evans and Rauch 1999). In turn, this better allows states to monitor, police, and collect revenue from citizens—to “see” the populace and thus control it (Scott 1998). 2 A capable bureaucracy is thus an important source of state strength, as it enables governments to administer and enforce enacted policy, afford and distribute public goods to redress popular grievances, and selectively crack down on violent dissenters (Goodwin 2001). Studies demonstrating the importance of state capacity for sophisticated policy implementation abound in the empirical literature; Perkins and Neumayer (2007) and Cole (2015), for example, find that states with limited bureaucratic capacity face more difficulties in implementing the stipulations of environmental and human rights treaties, while Cheon, Urpelainen, and Lackner (2013) show that low bureaucratic capacity makes governments more likely to employ gasoline subsidies, owing to the challenges they face in implementing less costly and inefficient means of income redistribution such as direct cash transfers. In the context of state–rebel interactions more specifically, Kalyvas (2006) argues that control of territory in civil wars enables actors to use selective rather than indiscriminate violence, by allowing them to collect information on individuals and activities in the area. The importance of state strength reflects a similar basic logic, as governments’ capacity for selective coercion will depend on the availability of intelligence on the identities of suspected terrorists, which will in turn depend on the quality of the state’s bureaucratic infrastructure and its ability to administer its territory (Scott 1998; Goodwin 2001; Fearon and Laitin 2003; DeRouen and Sobek 2004; Hendrix and Young 2014).
States with low levels of bureaucratic capacity, because they have limited administrative reach, will face difficulties in monitoring their populations—and, in turn, in collecting information on the identities of the terrorists, which would otherwise allow them to be more selective in their use of coercive violence. Thus, the implication for the provocation hypothesis is that low-capacity governments are more likely to be provoked. Since repression is a blunt and likely counterproductive means of responding, states are more likely to use repression if their other options are limited. An abundance of bureaucratic capacity should allow states alternative, more effective, and more fine-grained means of countering terrorism than repression, by enabling them to better collect and process intelligence on suspected terrorists, thus opening up the option for coupling selective coercion with amnesty for defectors (de Figueirado and Weingast 2001; Fearon and Laitin 2003; Kalyvas 2006). Empirically, Hendrix and Young (2014) find that terrorism is in fact more likely in states with powerful militaries than in states with high levels of bureaucratic capacity, a finding they attribute to high-bureaucratic capacity states’ ability to more effectively and selectively crack down on terrorists. Similarly, Goodwin (2001) argues that revolutionary movements are more likely to flourish in states with limited administrative capacity that rely on repression to defeat and discourage political opposition.
As discussed more below, regime type may also affect the likelihood of provocation, both by influencing governments’ ability to use force against their populations by shaping the legal constraints they face and by affecting their incentives for co-opting regime opponents by shaping their selectorates (Wilson and Piazza 2013). However, in light of the provocation logic of terrorism, repression will be suboptimal even for states free to use it, and even autocracies will have incentives to broadly reduce grievances in order to stave off violence or insurgency. State strength, in turn, conditions governments’ ability not only to collect information and selectively coerce but also to collect revenue and disburse benefits to citizens in order to discourage political violence (Fearon and Laitin 2003; DeRouen and Sobek 2004; Collier and Hoeffler 2004).
In Peru, for example, which is coded as having very low bureaucratic capacity, the government extensively violated the physical integrity rights of its populace during the 1980s and 1990s, put the armed forces in charge of its counterinsurgency and counterterrorism efforts, and declared forty-six states of emergency in response to the Sendero Luminoso’s terrorist campaign. By contrast, the counterterror efforts of higher-capacity Spain against the Euskadi Ta Askatasuna (ETA) during the late 1970s and early 1980s did not involve a campaign of indiscriminate repression. On the contrary, the Spanish response was much more judicious and included efforts to redress grievances—including increased autonomy for the Basques and general increases in civil liberties—as well as provisions for targeted amnesty to potential ETA defectors (Holmes 2001). 3 Moreover, Peru was a democracy during the period, and thus its repressive campaign cannot be explained by a lack of constraints on the government authority.
A variety of existing theories, then, suggest that high-capacity and low-capacity states should in fact react differently to terrorism. In particular, I argue, they should differ in their propensity to respond to terrorism with repression. Yet this relationship has not yet been empirically verified. Since repression is a suboptimal means of dealing with the threat of terrorism, and since bureaucratic capacity captures governments’ ability to collect information on, collect revenue from, and control the population, I expect states with high bureaucratic capacity to avoid blunt repression and instead use more selective means of cracking down. Moreover, bureaucratic capacity also facilitates governments’ ability to collect revenue and disburse public goods as a means of reducing the grievances that may be fueling rebel groups’ use of violence.
In this light, I argue that a dearth of bureaucratic capacity will limit states’ counterterrorism options such that governments may see little choice but to use repression.
Alternative Explanation: Regime Type
Alternatively, studies on the connection between regime type and terrorism point to the role of institutional constraints and public opinion in conditioning states’ responses to terrorist incidents. Democracies, however, will be prone to contradictory impulses in counterterrorist policy. On the one hand, public opinion is often galvanized by terrorist attacks, and policy makers are likely to feel pressured to retaliate, given that they are ultimately responsible to the citizenry that is being targeted by the terrorists (Davis and Silver 2004; Kydd and Walter 2006; Kibris 2011; Perliger 2012; Conrad, Conrad, and Young 2014a; Conrad et al. 2016). On the other hand, many studies have found that democracies actually experience more terrorism than nondemocracies, with one of the most prominent explanations being that democratic states, because they are both more constrained institutionally and more accountable to their people, face difficulties in responding with repression to terrorist attacks (Eubank and Weinberg 1994; Li 2005; Young and Dugan 2011; Wilson and Piazza 2013; San-Akca 2014). As Stanton (2013) and Pape (2003) point out, democracies may be more sensitive to civilian casualties but that could just as easily lead to concessions as to violence and repression.
Democracies, then, are likely to be constrained in the extent to which they can repressively respond to terrorism. This is due to both institutional constraints on the government via checks and balances and to democracies’ accountability to the people that they would be repressing. Indeed, Perliger (2012) finds that strong, established democracies are less likely to use “hard-line” measures in response to terrorism than are weaker, more partial democracies—although other studies find that coercive responses are far from absent, even among the former (Kibris 2011; Conrad, Conrad, and Young 2014a; Conrad et al. 2016). Moreover, Daxecker and Hess (2013) show that repression is generally more ineffective in combating terrorism in democracies than in autocracies. Thus, while there is no guarantee that democracies will not crack down, owing to public pressures for action, a large body of literature suggests that constraints on executive power will reduce the probability of repression.
While Hypothesis 3 does not address the fundamental puzzle of provocation, it does suggest that fewer constraints on the executive will make it less difficult for governments to employ repression. Ultimately, I argue that state capacity provides the most complete explanation for why states might be provoked by terrorism, as it explains why more preferable means of counterterrorism may be unavailable. It suggests that states are only likely to use repression when they have no other options for combating the threat of terrorism—namely, when their bureaucratic capacity is such that more selective means are rendered comparatively more challenging by the government’s inability to monitor the population, collect intelligence on potential terrorists, and selectively crack down on them.
Research Design
This article uses a cross-national analysis of all possible countries, a maximum of around 160, to evaluate the hypotheses enumerated above. The unit of analysis is the country-year, and the sample includes all countries for which data are available between 1981 and 2011, as this is the time period for which Cingranelli and Richards (CIRI) human rights data exist. I use time series ordinal logistic regression, with random effects to account for unit heterogeneity. A Wald test indicates the presence of panel heteroscedasticity (Greene 2000); thus, robust standard errors clustered by country are employed (Arellano 1987).
Dependent Variables (DVs)
In order to proxy for state repression, I use both the Political Terror Scale (PTS), which draws on the US Department of State’s Country Reports on Human Rights Practices and Amnesty International’s Annual Report to create an ordinal measure (1 through 5) of political terror, 4 and the CIRI Human Rights Data Project’s score for physical integrity violations (PHYSINT; Gibney et al. 2013; Cingranelli, Richards, and Clay 2013). The use of two measures allows me to ensure that the findings are not artifacts of one operational definition. For the PTS score, I use the average of the State Department and Amnesty International scores, when both are available; or whichever one is available, when both are not. The four component measures of PHYSINT—torture, political imprisonment, extrajudicial killing, and disappearances—are scaled 0–2, with 0 indicating that these occur frequently and 2 indicating that they occur rarely if ever. I rescale this variable so that higher numbers indicate lower physical integrity in order to facilitate interpretation of the results (such that larger, positive coefficients indicate more repression) and to make the findings using PHYSINT more comparable to those using PTS.
Independent Variables
To operationalize terrorism, I include both incidents of international terrorism in addition to incidents of domestic terrorism. Provocation theory implies that domestic crackdowns are the mechanism by which government overreaction can turn the tables in favor of the terrorists; however, it is possible that governments could violently overreact in response to attacks undertaken by foreign nationals as well. I use both the total number of incidents of terrorism in a year and the number of casualties (fatalities and injuries) from terrorist attacks in a year. 5 I obtain data on terrorism from the Global Terrorism Database (GTD; National Consortium for the Study of Terrorism and Responses to Terrorism, 2012). 6 I include all domestic and foreign incidents for which there is no doubt that the incident was an act of terrorism and which fulfills the GTD’s inclusion criteria. I also run analyses in which I examine domestic and transnational terrorism separately.
Furthermore, in order to test Hypothesis 2, I interact terrorism incidents with bureaucratic capacity, as states that are capable of responding with more targeted means are less likely to choose repression. Bureaucratic capacity can be measured in a number of ways, such as gross domestic product (GDP) per capita, taxation as a percentage of GDP, or the International Country Risk Guide’s (ICRG) measure of states’ “bureaucratic quality” (Hendrix 2010; Hendrix and Young 2014), which is available since 1985 for 141 countries. Since GDP per capita captures economic development more closely than state capacity, and taxation and expenditures data are only widely available from the late 1990s, I use bureaucratic quality to capture bureaucratic capacity, with data coming from the ICRG (Coplin, O’Leary, and Sealy 2007). 7 Importantly, this measure is independent of the prevalence of terrorism in a given country; rather than capturing internal instability, as other ICRG political risk variables do, the bureaucratic quality measure captures whether “the bureaucracy has the strength and expertise to govern without drastic changes in policy or interruptions in government services,” and whether “the bureaucracy tends to be somewhat autonomous from political pressure and to have an established mechanism for recruitment and training” (Howell 2011, 7).
For Hypothesis 3, I also interact terrorism with states’ level of constraints on executive power. Coding regime type using a composite measure of democracy, such as the Polity score, could be problematic, since the very factors that go into democracy scores may themselves be influenced by the repression that is taking place to squelch the rebels. 8 In this case, the operational definitions of these dependent and independent variables might be correlated by construction. Lagging the regime type score is a partial solution; however, to safely avoid tautology, I use states’ executive constraints (XCONST) score, a component of the Polity score which measures institutional limits on executive power and which comes closest to signifying what scholars pointing to the institutional limitations on democracies in this literature have described (Vreeland 2008).
Control Variables
In addition to regime type, the literature on repression suggests that a number of additional factors have an important impact on governmental PHYSINT which should be controlled for. First, I control for GDP per capita (in constant 2005 US dollars), with data from the World Development Indicators (2014). This provides an indicator for states’ level of economic development, as grievances may be fewer in wealthier countries such that there is less need for the government to use coercive force as a means of suppressing domestic threats that might otherwise arise from deprivation (Poe and Tate 1994; Poe, Tate, and Keith 1999; Keith 1999; Piazza and Walsh 2009; Robison 2009; Shor et al. 2014). Although GDP per capita is correlated with bureaucratic quality, leaving out the former is likely to produce omitted variable bias since wealth might both help states produce bureaucratic capacity, insofar as governments will then have more resources with which to build their institutions, and have an independent effect on the likelihood of state repression (as argued by the repression literature). As a robustness check, I run the analysis of Hypothesis 2 after dropping GDP per capita, and the results remain unchanged.
Next, I control for the presence of ongoing interstate wars and civil wars. Repression may be more likely during an interstate war as a means of clamping down on domestic dissent. Similarly, civil wars represent a direct challenge to the legitimacy and survival of the government in power, and thus states may turn to repression as a means of preserving their regimes (Poe and Tate 1994; Poe, Tate, and Keith 1999; Davenport 2007; Piazza and Walsh 2009). Both of these variables are dummy indicators, where 1 indicates the presence of an ongoing war. Data are obtained from the Uppsala Conflict Data Program/Peace Research Institute Oslo Armed Conflict Dataset (Gleditsch et al. 2002; Themnér and Wallensteen 2014).
Third, I control for logged population, as a larger population not only increases the number of opportunities for repression by simple probability but can also strain national resources in ways that produce popular dissatisfaction and, in turn, repression (Poe and Tate 1994; Poe, Tate, and Keith 1999; Keith 1999; Piazza and Walsh 2009; Dreher, Gassebner, and Siemers 2010). Data come from the World Development Indicators. Fourth, I follow Piazza and Walsh (2009) in controlling for whether a state has ratified the Optional Protocol to the International Covenant on Civil and Political Rights (ICCPR), as ratification has been shown to be negatively associated with repression (Keith 1999). All covariates are lagged by one year in order to ameliorate the potential for reverse causality, and because there may be a delay in the implementation of repression policies.
Fifth, I include a lagged DV, as the repression indicators are slow moving, and a number of scholars have found that repression is highly path dependent (Poe and Tate 1994; Davenport 1995, 1996; Poe, Tate, and Keith 1999). By including a lagged DV, one can be more sure that the previous year’s terrorism actually had an effect on repression, rather than the current year’s value being solely a holdover from the previous year’s level. This also serves to curb the potential problem of serial correlation. Finally, I include year dummies in order to account for temporal trends or year-specific shocks.
The regression equation used to test Hypothesis 2 is as follows:
where i indexes countries and t indexes years, X i,t−1 represents a vector of country-specific economic and political control variables, ζ t is a vector of year fixed effects, and ε i,t is a stochastic error term. Repression represents either the PTS or CIRI’s PHYSINT index, while incidents represents the number of terrorist incidents a country experienced in a given year and Bureaucratic quality refers to the ICRG’s country-specific score of bureaucratic strength and capacity.
Results
Baseline Results
The results before adding interaction terms are presented in Table 1. 9 As expected, terrorist incidents, the presence of an ongoing civil war, and higher population sizes all increase the likelihood of state repression, while economic development and XCONST decrease the likelihood, and ICCPR ratification and interstate wars have no effect. The results using casualties from terrorism instead of the number of incidents are similarly positive, but the effect is of a smaller magnitude—each terrorist incident increases the odds of being in a higher category of PHYSINT by 0.59 percent and of political terror by 0.7 percent, while each casualty increases the odds of both by 0.04 percent. 10 Thus, consistent with previous studies (Piazza and Walsh 2009), I find support for Hypothesis 1 and confirm the baseline relationship between terrorism and repression; terrorist incidents and casualties inflicted by terrorist attacks increase the likelihood of higher levels of political terror and PHYSINT.
Baseline Results: Terrorist Incidents and Casualties.
Note: Values represent ordinal logistic coefficients. Robust standard errors in parentheses. All models include year dummies. GDP pc = gross domestic product per capita; ICCPR = International Covenant on Civil and Political Rights; PHYSINT = physical integrity violations; PTS = Political Terror Scale.
*p < .05.
**p < .01.
***p < .001.
Interaction Effects
Hypotheses 2 and 3 suggest an interactive relationship between terrorism, on the one hand, and state capacity and XCONST, on the other. As such, in Tables 2 and 3, I add a variety of interaction terms to the baseline results. The addition of these terms yields results that suggest conditions under which states are provoked by acts of terrorism. However, since interpretation of these terms is difficult, I report the marginal effects of these interactions, which indicate the impact of terrorist incidents on repression given values of these moderating variables, in Figures 1 and 2.
Interaction between Terrorism and State Capacity.
Note: Values represent ordinal logistic coefficients. Robust standard errors in parentheses. All models include year dummies. GDP pc = gross domestic product per capita; ICCPR = International Covenant on Civil and Political Rights; PHYSINT = physical integrity violations; PTS = Political Terror Scale.
*p < .05.
**p < .01.
***p < .001.
Interaction between Terrorism and Executive Constraints.
Note: Values represent ordinal logistic coefficients. Robust standard errors in parentheses. All models include year dummies. GDP pc = gross domestic product per capita; ICCPR = International Covenant on Civil and Political Rights; PHYSINT = physical integrity violations; PTS = Political Terror Scale.
*p < .05.
**p < .01.
***p < .001.
Terrorism’s effect on repression, conditioned by state capacity.
Terrorism’s effect on repression, conditioned by executive constraints.
Turning first to the impact of terrorism on repression given bureaucratic quality (Figure 1), these effects show that the relationship is strongest in countries with lower levels of state capacity and that the effect moves toward zero at higher levels of state capacity. Although there is some overlap in the confidence intervals, namely, when PHYSINT are used as the DV, there is a clear negative trend in the estimates. This is in line with the expectations of Hypothesis 2 and suggests that it is low-capacity states which are most likely to use repression, because their ability to use more selective means is limited.
Finally, regarding XCONST as a moderator (Hypothesis 3), these results suggest that XCONST have little conditional effect for whether governments use repression in response to terrorism. As shown in Figure 2, increased repression is marginally more likely in states with greater limits on executive authority than in states with fewer limits—although the confidence intervals overlap entirely and the coefficients are quite flat, suggesting a null effect.
Endogeneity
Endogeneity could be problematic here, as it is possible that repression could cause terrorism rather than the reverse (Walsh and Piazza 2010). Thus, like Dreher, Gassebner, and Siemers (2010) and Shor et al. (2014), I employ Arellano–Bond dynamic panel data estimation to replicate the original results. I use the lagged DVs, terrorist incidents, and the interactions terms as the endogenous regressors, while the other controls are used as exogenous variables. In order to reduce the otherwise enormous amount of instruments, I collapse the matrix of instruments to a single column, resulting in a maximum number of instruments of 96, before adding interactions, and 127 including interactions (Roodman 2009). An Arellano–Bond test for autocorrelation indicates the presence of first-order but not second-order serial correlation for all models, while a Sargen–Hansen test does not reject the null hypothesis that the overidentifying restrictions are valid. This suggests that the instruments are valid and that the results are reliable.
The results, presented in Tables 4 and 5 with marginal effects in Figure 3, are largely consistent with the original findings. Unlike ordinal logistic regression, the Arellano–Bond estimator assumes a continuous DV, and so these results can be interpreted directly as the change in the DV, given a one-unit change in each independent variable. In the baseline results, the terrorist incidents variable remains positive in direction, although it only achieves statistical significance at the .05 level, when PTS is used as the DV. However, the marginal effects are very consistent with the initial results. Bureaucratic quality still has a highly negative conditional effect—indeed, the negative trend in the conditional effects is even more pronounced—while XCONST have essentially no effect. Thus, as before, terrorist incidents have a positive effect on repression, supporting Hypothesis 1, while Hypothesis 2 receives strong support and Hypothesis 3 is not supported.
Arellano–Bond Dynamic Panel Data Models, One-step Difference Generalized Method of Moments (GMM), and Physical Integrity Violations (PHYSINT).
Note: Robust standard errors in parentheses. All models include year dummies. GDP pc = gross domestic product per capita; ICCPR = International Covenant on Civil and Political Rights.
*p < .05.
**p < .01.
***p < .001.
Arellano–Bond Dynamic Panel Data Models, One-step Difference Generalized Method of Moments (GMM), and Political Terror.
Note: Robust standard errors in parentheses. All models include year dummies. GDP pc = gross domestic product per capita; ICCPR = International Covenant on Civil and Political Rights; PTS = Political Terror Scale.
*p < .05.
**p < .01.
***p < .001.
Marginal effects, Arellano–Bond generalized method of moments models.
Domestic versus Transnational Terrorism
Since most prior studies have focused on the effect of transnational terrorism on repression (Piazza and Walsh 2009; Robison 2009; Dreher, Gassebner, and Siemers 2010), I also separate terrorist incidents into those that are made by domestic actors and those that are made by foreign actors. 11 The results, listed in the Online Appendix, show that when used separately, both domestic and transnational terrorist incidents have a significant and positive effect on state repression, using both PHYSINT and political terror as DVs. The coefficient on the transnational terrorism variable is larger than that of domestic terrorism. However, when both domestic and transnational incidents are controlled for together, the transnational terrorism variable loses its statistical significance. The null result, although, is somewhat offset by the use of marginal effects for the interaction terms. 12 State capacity has a negative conditional effect on the relationship between both domestic and transnational terrorism, as expected by Hypothesis 2, although the difference between high-capacity and low-capacity states is much greater for transnational terrorism.
As for XCONST, I find that constraints on executive authority have little influence on governments’ likelihood to repress in response to both domestic and transnational terrorism. Before controlling for the effects of domestic terrorism, countries with higher constraints on executive power are shown to be slightly less likely to repress in response to transnational terrorism and slightly more to repress after domestic terrorism. However, the overlap in the confidence intervals between the highest and lowest values of XCONST calls this relationship into doubt. Moreover, the trend further flattens once both types of terrorism have been controlled for.
To summarize, then, high-capacity states are particularly less likely to repress in response to transnational terrorism than are low-capacity states, while the results suggest that XCONST do not have a strong conditional effect on the probability that governments will use repression after terrorist incidents.
Additional Robustness Check 13
To test the robustness of these findings, I adopt a number of measures, such as adding new controls and changing the operationalization of some variables. First, I interact the terrorist incidents variable with the number of foreign policy rivalries each country faces. States’ external environment may impact how they perceive and react to terrorist attacks; however, the question of how states balance internal and external threats and priorities has attracted valuable but limited attention in the international relations literature more broadly. Existing studies have tended to argue that states—in particular fragile states in the developing world—often prioritize internal threats if their capabilities are limited (David 1991; Barnett and Levy 1991). Indeed, as Davies (2016) notes, repression can divert needed resources and manpower from being used to deter or defend against international adversaries. Rivals can also act as state sponsors for terrorist groups, which may make counterterrorism efforts one and the same with the pursuit of external security (Conrad 2011). I obtain data on international rivalries, which extend until 2001, from Klein, Goertz, and Diehl (2006).
The baseline results hold up even after controlling for states’ number of rivalries. The rivalry variable, however, only has a statistically significant effect on repression, when PTS is used as the DV. This result is reflected in the marginal effects as well. Rivalries barely affect the relationship between terrorism and PHYSINT, but they do have a limiting effect on the relationship between terrorism and political terror (using PTS). Terrorist incidents increase the likelihood of political terror in countries with few rivalries, but this relationship declines and ultimately disappears for states with six or more rivalries.
When enduring rivalries (those lasting at least twenty years) are used instead of all rivalries, the results become much more decisive. Using both PHYSINT and PTS as the DV, the marginal effects show that increasing the level of enduring rivalries diminishes the effect of terrorist incidents on repression. Thus, these results suggest that states’ foreign policy environments condition their propensity to repress in response to terrorism.
Second, the results largely hold up when the composite Polity score is used in place of the XCONST score. The relationship between terrorist incidents and repression remains near constant across all levels of the Polity score. Following Piazza and Walsh (2009), I also interact terrorism with a dummy variable for anocracies (those regimes falling between −5 and 5, inclusive, on the Polity scale) to assess whether they behave differently from democracies and autocracies, as anocracies are often less stable and more prone to violate human rights (Fein 1995; Vreeland 2008). The results show that anocracies do not behave uniquely, as the coefficient on the interaction term does not reach statistical significance using either PHYSINT or PTS as the DV. Thus, the results again indicate that regime type ultimately matters much less than state capacity for whether states are provoked by terrorist incidents. However, as noted previously, since the Polity score and repression may be tautologically related, these results are somewhat suspect.
Third, I add additional control variables. The first of these is the percentage of a state’s population that resides in urban areas, as urbanization might be related to state repression because it provides a rough proxy for how “modernized” countries are, which in turn may be linked to the availability of other, less blunt means of dealing with dissent than repression, and to the emergence of norms of nonviolence (Shor et al. 2014). Data come from the World Development Indicators (2014). The second is a dummy variable indicating whether the country is an Organization for Economic Cooperation and Development member, as the advanced economies generally exhibit better human rights behavior (Dreher, Gassebner, and Siemers 2010). Third is a dummy variable indicating whether one or two PTS values are available. The results are consistent with the initial findings; terrorism continues to have a positive effect on the probability of repression in the baseline results, while bureaucratic capacity still has a negative conditional effect and XCONST has a negligible effect.
Next, although using too few controls runs the risk of omitted variable bias, the inclusion of too many control variables can also bias the estimates or increase the standard errors (Achen 2002; Clarke 2005). In particular, the use of a lagged DV, which is almost always highly significant and large in magnitude in time-series models, runs the risk of reducing the estimates of the other covariates (Achen 2000). Thus, I replicate the initial results, dropping the lagged DV. As expected, the terrorism variables increase in magnitude and have lower p values. However, the results are highly consistent with initial findings; bureaucratic capacity continues to be the most decisive moderator of the relationship between terrorism and repression, while XCONST have little conditional effect. Additionally, in order to ensure that the year dummies do not soak up so much variation in repression that the estimates of the other independent variables are imprecise (Greene 2004), I replace the dummies with linear, squared, and cubic time trends. The new results are virtually identical.
Finally, I run the analysis using country fixed effects and linear, squared, and cubic time trends in place of random effects and year fixed effects, since the coefficients using random effects may be biased if there is correlation between the independent variable and the unobserved unit effects (Clark and Linzer 2015). I do so both while still including the lagged DV and after dropping it, since the use of a lagged DV may induce bias when it is used in combination with fixed effects (Nickell 1981). Moreover, I also run the analysis using country fixed effects and year fixed effects together. The results are highly robust; terrorist incidents and casualties increase the probability of repression, while bureaucratic capacity continues to negatively influence governments’ propensity to use repression in response to terrorism, and XCONST have minimal effect.
Conclusion
In sum, these results suggest that states are indeed provoked by terrorist attacks in some cases. The expectations of Hypothesis 1 and Hypothesis 2 are borne out by the evidence, while Hypothesis 3 is not supported. Terrorism is associated with greater repression in low-capacity states much more than in high-capacity states, while XCONST have a negligible effect. More broadly, terrorist incidents and casualties from terrorism are, all else held constant, associated with more political repression. Moreover, I find preliminary evidence that countries facing many foreign policy rivals respond less repressively to terrorism than states with fewer foreign enemies. This suggests that states do face trade-offs in managing internal and external security threats and that multiple threats force states to prioritize some at the expense of others.
In terms of the broader theoretical expectations discussed above, bureaucratic capacity is shown to have the strongest and most consistent conditional effect on the likelihood of provocation. This provides further evidence that state capacity is an important determinant of whether states undertake potentially ineffective and even counterproductive policy measures, such as blunt repression. Moreover, this study suggests another potential mechanism by which low state capacity might relate to the causes of terrorism: by limiting governments’ options for counterterrorism, thus rendering them more vulnerable to falling into the provocation “trap” of using repression as a means of counterterror and, in turn, encouraging dissidents to use terrorism.
Regarding the provocation logic of terrorism more specifically, since provocation is arguably the strategy of terrorism that most heavily depends on how governments choose to react, this study suggests that it is important to qualify the theory by making note of the fundamental puzzle as to why governments would fall into the trap of using repression (Kydd and Walter 2006; Kalyvas 2006). Moreover, to address this puzzle, one should take into account factors that limit the government’s ability to respond in other, less counterproductive ways—in particular, bureaucratic state capacity.
A potential avenue for future research on the conditions under which states are provoked would be to investigate whether the objectives of terrorist groups emphasized provocation and thus to try to separate cases of terrorism actually aimed to evoke a government overreaction from cases in which the objective was attrition, intimidation, or spoiling, to name just a few (Kydd and Walter 2006). Identifying cases of the former would permit for more fine-grained comparison by granting researchers a more comparable sample of incidents from which inferences on government propensity to respond with repression could be made. Additionally, further theorizing on the conditions under which countries’ populations are likely to blame the government instead of the terrorists for any repression that occurs would provide even more insights about when governments are likely to use coercive counterterrorist measures.
One might also consider alternative means of operationalizing “provocation.” Political repression is certainly a reasonable proxy, given that provocation theory expects states to overreact in costly and potentially self-defeating ways; however, it may also be a fairly noisy measure. Promising data collection efforts by Epifanio (2011) and Carter (2014) seek to more directly measure state responsiveness to acts of terror by specifically examining countries’ legislative responses and the counterterrorist activities of their armed forces and police, although they are limited in scope to the West.
Footnotes
Acknowledgment
I would like to thank Page Fortna, Jack Snyder, Johannes Urpelainen, Shawn Lonergan, Nick Lotito, Renanah Miles, Asli Saygili, and two anonymous reviewers for helpful feedback on earlier drafts of this article. A previous version of this article was presented at the 2014 annual ISAC-ISSS Conference, for which I gratefully acknowledge travel funding from the Columbia University Department of Political Science and the Columbia University Graduate School of Arts and Sciences. Any errors are my own.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
