Repression and Terrorist Group Duration in Democracies

We expect that adopting harsh techniques in response to terrorist events results in backlash effects in democracies. While repressive measures are intended to weaken a terrorist group and compromise its security, such strategies likely backfire in regimes that derive their support from the guarantee of civil rights and liberties. Since the existence of civil rights and liberties is widely accepted in democracies, harsh government responses to terrorism – such as holding suspects without charging them, assassinating suspected terrorists, curbing civil freedoms or imposing retribution on alleged sponsors – thus contradict the fundamental values of such regimes. As Crenshaw points out, governments risk becoming ‘the victims of their own stereotype of terrorism as responsive only to force’.Footnote ¹⁸ Consequently, government crackdowns on extremists, and accompanying policies that limit political freedoms and rights, undermine the government's legitimacy and can increase support for terrorist groups, facilitate terrorist recruitment, and prolong the lifespan of terrorist groups.Footnote ¹⁹

Such backlashes are particularly likely if counter-terrorist measures do not discriminate between supporters of terrorist organizations and innocent citizens.Footnote ²⁰ Proactive measures may thus increase popular sympathies for terrorist groups’ grievances, and could even boost terrorist recruitment. As Rosendorff and Sandler point out, coercive tactics ‘may have a downside by creating more grievances in reaction to heavy-handed tactics or unintended collateral damage’.Footnote ²¹ We propose three mechanisms to explain why elites in democracies experience backlash effects.Footnote ²²

First, we argue that the use of harsh policies reduces local communities’ co-operation with government authorities. Since terrorism is a relatively rare and dispersed form of violence, effective counter-terrorism policing depends on complete, accurate and well-timed information from local communities.Footnote ²³ However, intrusive government actions in response to terrorist threats can undermine government legitimacy and estrange local communities from government authorities.Footnote ²⁴ Hence, government responses that are perceived as harsh and excessive can lead individuals to withhold information out of fear of unfair treatment. Surveys of Muslims in the United States and the United Kingdom show that their willingness to alert the police to threats is ‘negatively affected by the belief that Muslims are subject to discrimination’.Footnote ²⁵ Research has also documented that perceptions of intrusive police tactics against Muslim Americans can reduce non-Muslims’ perceptions of the police as legitimate.Footnote ²⁶

Secondly, harsh counter-terrorism policies can increase popular sympathy for the group's cause and subsequently spur recruitment for terrorist organizations in democratic regimes.Footnote ²⁷ Coercive responses that undermine a government's legitimacy can thus increase: overall support for a terrorist group, the number of potential recruits and the resources available to the group.Footnote ²⁸ Accounts of the escalation of violence in several democracies illustrate this dynamic. Several studies of the Northern Ireland conflict emphasize that coercive government measures in the 1970s helped drive the growth and intensity of the IRA during this period.Footnote ²⁹ Similarly, Della Porta's analysis of counter-terrorism in Italy emphasizes how hard-line policies can produce counterproductive results by ‘stiffening individuals as regards their own choices and increasing solidarity outside’.Footnote ³⁰ In addition, harsh government tactics against terrorism in Israel have reportedly resulted in a large number of civilian casualties, which can fuel the cycle of violence by mobilizing new supporters and strengthening the motivation of group members.Footnote ³¹

Thirdly, we expect that democratic states’ use of repression will jeopardize their ability to combat terrorism more generally. Democratic leaders are accountable to the public and can be removed from office if their counter-terrorism policies are perceived as unnecessarily harsh and excessive. The use of coercive measures undermines the values that democratic regimes represent, and can thus threaten the survival of leaders. Leadership turnover may result in changes to counter-terrorism policies that can subsequently reduce the government's effectiveness in combating terrorist groups.Footnote ³² Moreover, simply being threatened with removal can produce inconsistent counter-terrorism policies, since democratic leaders may decide to ‘change policies midstream’ to avert popular criticism.Footnote ³³ In addition, Walsh and Piazza maintain that opposition parties will try to gain political capital by criticizing counter-terrorism policies, which will likely reduce the overall resources devoted to counter-terrorism.Footnote ³⁴

Repression and Terrorist Group Duration in Non-Democratic Regimes

We expect that coercive government action can successfully deter terrorist group survival in non-democratic regimes, because authoritarian leaders are less dependent on whether their counter-terrorism policies are perceived as legitimate.Footnote ³⁵ First, we do not expect repression to deter public co-operation in authoritarian regimes. Since repression is more common in non-democratic regimes, government legitimacy is already low. Research in the criminology literature shows that legitimacy becomes largely irrelevant for public co-operation when governmental abuse is considered an inevitable fact of life.Footnote ³⁶ Compliance in authoritarian regimes is thus likely to be driven by fear of punishment rather than legitimacy, so coercive counter-terrorism strategies are unlikely to further reduce community co-operation with the government.

Secondly, we anticipate that repressive measures by authoritarian regimes can successfully depress recruitment for terrorist groups. Gathering information on suspected terrorists and their sympathizers (and detaining them without trial) is unlikely to lead to backlash effects, because authoritarian governments can suppress information about their actions and curtail media reports.

Finally, leaders of authoritarian regimes are not as constrained in their counter-terrorism efforts as their democratic counterparts because they are less dependent on public opinion for remaining in office.Footnote ³⁷ Legal and normative constraints are subordinate in authoritarian regimes’ responses to anti-government violence.Footnote ³⁸ Authoritarian leaders can thus be less concerned with the public's response to repressive counter-terrorism policies or the electoral consequences of such actions.Footnote ³⁹

Therefore our main hypothesis expects the effects of repression on group duration to vary by regime type. Coercive measures in democracies constrain the civil rights and liberties expected by citizens and thus inadvertently increase the resilience of terrorist groups. Conversely, proactive measures are expected to be successful in deterring the use of terrorism and reducing the lifespan of terrorist organizations in non-democratic regimes.

Data

We evaluate our theoretical arguments in a statistical analysis of 539 terrorist groups in eighty-five countries for the 1976–2006 period.Footnote ⁴² We use data on terrorist organizations collected by the RAND-MIPT project and presented in Jones and Libicki. Our unit of analysis is the group-year; the last year of observation is 2006 (or earlier, if the group terminated prior to that date). The data include terrorist organizations that carried out international incidents before 1998 and organizations that carried out international or domestic incidents from 1998–2006. We include terrorist groups engaging in both domestic and transnational terrorism for several reasons. First, Jones and Libicki identify the host country (the country in which groups base their operations) rather than the target country (the country experiencing the violence) for terrorist organizations carrying out international incidents.Footnote ⁴³ Our empirical analysis investigates how government responses in the host country (in which groups base their operations) affect group duration, rather than government responses in the country that is targeted. Secondly, the RAND-MIPT database's definition of international terrorism includes many incidents committed by groups to which our theoretical argument should apply rather straightforwardly. Terrorist events are defined as international if terrorists go abroad to strike their targets, select victims or targets with foreign connections or attack airlines’ personnel or equipment.Footnote ⁴⁴ For attacks that occur within a host country and are thus in the natural environment of the group (the second and third categories of international terrorism), responses from that government should clearly be relevant for group duration. For example, the data include groups such as the African National Congress (ANC) because it engaged in attacks against international corporations in South Africa, yet the ANC's goal was arguably to establish racial equality in South Africa. Finally, we think our argument may also apply to groups that target internationally. While repressive government measures may not be aimed at groups that carry out attacks in other countries, coercion that affects large parts of the population would likely also affect the capabilities of such groups.Footnote ⁴⁵ A robustness test excluding groups that target internationally is presented in the online appendix and shows support for our main contention.

Dependent Variables

The first dependent variable is used to evaluate our hypothesis on group termination. It is a dummy variable coded 0 for each year that a particular group was in operation, and 1 for the year in which the group terminated. Groups still active in 2006 were coded as 0 in the last year of observation. Of the 539 groups studied, 322 groups ended their campaigns during 1976–2006. Data on start and end dates of terrorist organizations come from Jones and Libicki.Footnote ⁴⁶ Jones and Libicki used the following criteria to code the beginning and end of terrorist campaigns:

Jones and Libicki provide information on how terrorist campaigns end, which we used to create four additional dependent variables to examine the outcome of terrorist campaigns: government victory, political settlement, group victory and splintering.Footnote ⁴⁸Table 1 presents the distribution of terrorist campaign outcomes for the 322 groups that ended their campaigns in the period under analysis. The table shows that splinters, government victory and settlement are approximately equally common, whereas group victory is by far the least likely outcome of terrorist campaigns.

Table 1

Terrorist Campaign Outcomes, 1976–2006

Independent Variables

The data provided in Jones and Libicki include information on a number of group characteristics that we append with state-level characteristics, described in more detail below.Footnote ⁴⁹ The first independent variable measures the level of repression in states. Data for this variable were obtained from the Political Terror Scale (PTS).Footnote ⁵⁰ The variable is coded 1 through 5, ranging from least repressive to most repressive.Footnote ⁵¹ The second key independent variable measures how democratic countries are; it is taken from the Polity IV Project data.Footnote ⁵² The scale for this variable ranges from −10 (fully authoritarian) to +10 (fully democratic).Footnote ⁵³ To model the interaction between repression and regime type outlined in the first hypothesis, we create an interaction variable by multiplying the repression and democracy variables.Footnote ⁵⁴

Additional variables control for other group and state characteristics that likely affect group termination. Data for all group characteristics come from Jones and Libicki.Footnote ⁵⁵ First, group size has been shown to affect the likelihood of group termination in previous research: larger groups have greater longevity. The measure of the size of the terrorist group is an ordinal variable coded 1 for less than 100 people, 2 for 100–999, 3 for 1,000–9,999 and 4 for 10,000 or more.Footnote ⁵⁶

Secondly, the previous literature also identifies terrorist groups’ ideological views as a significant factor that affects group longevity; religious terrorist organizations are more durable than leftist or right-wing groups. We separate the ideology measures from Jones and Libicki into four dummy variables:Footnote ⁵⁷ nationalist groups, religious groups, leftist groups and right-wing groups.Footnote ⁵⁸ In our models, right-wing groups are the excluded category because they were least represented in the data.

Thirdly, we control for the breadth of terrorist goals. This measure is coded as an ordinal variable according to the scope of the groups’ goals.Footnote ⁵⁹ A fourth control variable incorporates information about the number of countries in which individual terrorist organizations operate or have bases of support. The foreign presence variable is a dummy coded 0 if the terrorist group is based in only one state and 1 if the group is present in two or more states.

Two additional variables account for other state characteristics. Gross domestic product (GDP) per capita – gathered from the Penn World TablesFootnote ⁶⁰ – serves as our measure of state economic development. Data for population size come from the World Bank's World Development Indicators.Footnote ⁶¹

Methodology

We use a Cox proportional hazards model to estimate the hazard of group termination, our first dependent variable. The model allows us to estimate the hazard function by predicting the failure or termination of conflict within a particular time interval. The hazard rate for any observation i at time t is given by the function h_i(t) = h₀(t) exp (x′_tβ), where h₀(t) is the baseline hazard of the event. In this type of model, hazard rates are only calculated as failures and are not sensitive to distribution issues because the baseline hazard rate cancels out when the hazard ratio is taken in the event of a failure. We use a stratified Cox model to analyse terrorist campaign outcomes, the second set of dependent variables in our empirical analyses.Footnote ⁶² This model lets us estimate the effect of covariates on a particular terrorist campaign outcome, rather than estimating the average effect of independent variables across different outcomes.Footnote ⁶³

One restriction imposed on the Cox model is the assumption of proportionality, meaning that the effect of the covariates on the hazard ratio has to be proportional during the lifespan of a terrorist group. Tests of the proportional hazards assumption indicated evidence of non-proportionality for several independent variables, in particular the group size variable.Footnote ⁶⁴ We follow Box-Steffensmeier and Jones’ advice and deal with time-varying covariates by interacting such variables with the logarithm of time.Footnote ⁶⁵

In the following tables for group duration and outcomes, the hazard ratio is a covariate's effect on the baseline hazard of group termination.Footnote ⁶⁶ Hazard ratios smaller than 1 indicate that an independent variable reduces the hazard rate of event occurrence within a specified time interval, and ratios larger than 1 mean that an independent variable increases the hazard rate of event occurrence. The statistical significance of hazard ratios is measured using two-tailed tests, and all models employ robust standard errors.Footnote ⁶⁷

Terrorist Campaign Duration

Table 2 presents the results for models of terrorist group duration. The first model in Table 2 includes all independent variables except the product term. In this model, findings for the polity variable mirror past studies that found no significant relationship between this indicator and terrorist group termination. Similarly, the hazard ratio for the repression variable is negative but insignificant. Previous large-sample studies have not included a measure for repression, and our findings suggest that repression does not affect group termination independently.

Table 2

The Determinants of Terrorist Group Termination

Note: Hazard ratios with robust standard errors in parentheses are reported.

**p < 0.01; *p < 0.05 (two-tailed tests)

Among the control variables, the coefficient for population is negative and significant, indicating that the larger the population of a particular country, the less likely it is that terrorist groups will terminate. The variable for group size is significant and negative, which indicates that larger groups are less likely to fail. Nationalist terrorist groups are less likely to terminate than right-wing terror groups (the excluded category). With regard to terrorist group goals, we find that groups with more extensive aims are less likely to end their campaigns than groups pursuing less ambitious goals. Coefficients for the GDP per capita variable, leftist groups, religious groups and the variable for a group's foreign presence are not significant.

The second model in Table 2 includes the interaction term. The hazard ratio is significant and negative (z = −2.69), indicating that repression is a counterproductive strategy in democracies because it decreases the risk of terrorist group termination, but increases the hazard of termination in authoritarian regimes. This finding supports Hypothesis 1, which expected that the effect of repression varies with regime type. The results for the control variables are similar to the first model. The third model in Table 2 presents the results for a parametric model with a Weibull hazard function. While the Weibull model is more restrictive than the non-parametric Cox model in assuming that the underlying shape parameter is known, the results show that our findings are robust to alternative estimation techniques.Footnote ⁶⁸ The hazard ratio for the interaction is negative and significant. The control variables are similar to the first two models in Table 2, although we find that religious groups (rather than nationalist groups) are less likely to terminate than right-wing organizations in the Weibull model.

Table 3 presents substantive effects for all models in Table 2. We first discuss substantive effects for the interaction in the Cox model. When the polity variable is set at 10, the repression variable is varied from 1.5 to 4.5 and the interaction is varied along with the respective values of the constitutive terms, the hazard of group termination decreases by 33 per cent. Conversely, repression increases the likelihood of terrorist group failure in non-democratic regimes. Setting the polity variable at −3, varying the repression variable from 1.5 to 4.5 and varying the interaction along with the respective values of the democracy and repression variables produces a 76 per cent increase in the hazard rate of termination. Substantive effects for the Weibull model are more meaningful to interpret, as they indicate the change in the expected duration of terrorist groups. For democracies, the expected duration of terrorist groups increases from thirteen to twenty-one years if the repression variable and the interaction are varied from 1.5 to 4.5. Conversely, the duration of groups in authoritarian regimes decreases from twenty-five to thirteen years if the repression variable is varied from 1.5 to 4.5, together with the interaction.

Table 3

The Determinants of Terrorist Group Termination (Substantive Effects)

^a Polity is set at 10 (+1 SD from the mean) and repression is varied from 1.5 to 4.5 ( ±1 SD from the mean) together with the interaction to calculate marginal effects on the hazard rate in democracies.

^b Polity is set at −3 (−1 SD from the mean), repression is varied from 1.5 to 4.5 ( ±1 SD) together with the interaction to calculate marginal effects on the hazard rate in non-democracies.

Figure 1 is based on the second model in Table 2 and illustrates the effect of repression on the likelihood of failure in democratic and non-democratic states.Footnote ⁶⁹ In this figure, the first graph depicts the effect of repression on group termination when the democracy variable is set at 10 (+1 SD from the mean) and the second graph presents the same effect when the democracy variable is set at −3 (−1 SD from the mean). In both graphs, repression is varied from 1.5 to 4.5 (±1 SD from the mean), and the interaction is varied along with the respective values of the constitutive terms. The first graph in Figure 1 illustrates the effect of repression on group termination in democratic states. High repression in democracies, as represented by the solid line, results in lower rates of termination than low levels of repression, as represented by the dotted trend line. The opposite effect is observed in the second graph shown in Figure 1. For non-democracies, higher repression (solid line) leads to substantially higher cumulative hazard ratios and therefore a greater likelihood of terrorist group termination over time than low repression (dotted line). Taken together, the results presented in Tables 2 and 3 and illustrated in the figure support our expectations.

Fig. 1

The effect of repression on terrorist group duration

Terrorist Campaign Outcomes

Next we evaluate how repression and democracy influence the outcome of terrorist campaigns. Two caveats are necessary before we discuss our empirical results presented in Table 4. First, the results for our outcome models are based on a small number of data points, in particular for campaigns ending with a terrorist group victory (an outcome observed in less than 7 per cent of campaigns that ended). Secondly, it may be difficult to determine, in practice, whether an outcome was in fact favourable to the government or the group. A government victory, for example, may come only after a protracted and costly counter-terrorism campaign, such as the Sri Lankan government's defeat of the Tamil Tigers in 2009.

Table 4

The Determinants of Terrorist Campaign Outcomes

Note: Reported are hazard ratios with robust standard errors in parentheses.

**p < 0.01 *p < 0.05 †p < 0.1 (two-tailed tests).

The second hypothesis expects that repression in democracies reduces the likelihood of government victory and settlements, outcomes that are more favourable for governments. We find partial support for this hypothesis. The hazard ratio for the interaction between the repression and polity variables is negative in both the government victory and settlement models, but reaches statistical significance only for terrorist campaigns that resulted in political settlements (z = −1.99). While we expected that government coercion would be counterproductive for all outcomes favourable for the government, it is interesting to observe that repression in democracies appears particularly damaging for outcomes in which governments depend on groups’ willingness to enter negotiations and come to a mutual agreement. Coercive responses thus appear to be less influential for outcomes in which governments are capable of imposing their preferred outcome.

The model for group victory shows support for Hypothesis 3. The hazard ratio for the interaction is positive and significant (z = 2.20), which demonstrates that repressive measures increase the risk of group victory in democracies, but reduce the hazard of group victory in authoritarian regimes. As expected in Hypothesis 3, we find that the effect of coercive responses is reversed for outcomes that are unfavourable for governments. Our findings thus show the importance of distinguishing among different outcomes rather than collapsing all groups that ended their campaigns into a single category. As expected, we do not find a significant relationship between repression and regime type for groups that end in splinters.

Some interesting findings emerge for other covariates in the outcome models. Population size increases the likelihood of a government victory, but reduces the likelihood of settlements or group victories. Findings for the group size variable show that campaigns fought by larger groups are less likely to end with government victories, settlements or splinters, but do not affect the hazard of group victories. Compared to right-wing groups, leftist, religious and nationalist groups are less likely to negotiate settlements to their campaigns. Yet we also find that religious groups are less likely to defeat the government and more likely to splinter. The breadth of groups’ goals reduces the hazard of settlement and group victory and increases the hazard of splintering, but has no effect on government victory. We do not find evidence that GDP or foreign bases affect terrorist campaign outcomes.

Robustness Tests

We conducted a large number of additional robustness tests to evaluate whether findings for terrorist group duration are robust to the inclusion of potentially influential observations, additional operationalizations of key independent variables, alternative samples of terrorist organizations and terrorist attacks as the dependent variable. The results for robustness tests are described in detail in the online appendix. Statistical models that exclude splinter groups, groups that target internationally and insurgent groups confirm that repression increases group duration in democracies but lengthens terrorist campaigns in authoritarian regimes. The results also remain consistent when we use Cingranelli and Richards’ measure on physical integrity rights.Footnote ⁷⁰ An additional robustness check disaggregates authoritarian regimes and shows that personalist regimes are less susceptible to backlash effects when using repression than other types of authoritarian regimes. We also find that the conditional effects of repression and regime type on group duration are supported for two alternative samples of terrorist groups. Findings using data on terrorist groups from Asal and Rethemeyer and the Global Terrorism Database (GTD) are largely consistent with earlier results.Footnote ⁷¹ Finally, a robustness check using terrorist attacks from the GTD as the dependent variable demonstrates that the implications of our argument hold for the incidence of terrorism: repression in democracies increases the likelihood of terrorist attacks, but reduces the probability of incidents in authoritarian regimes.