Scholars have recently become more interested in the relationship between democracy and terrorist violence, but increased research on the topic has still left several important questions unanswered. For instance, many studies produce evidence that the presence of democratic institutions increases the likelihood of terrorist targeting,Footnote 1 but others show that democracy diminishes the prospects of such violence.Footnote 2 Interestingly, contributors to both sides have used the same democratic attributes to develop their theoretical arguments. In particular, the protection of civil liberties by most liberal democracies allows activist groups within a society to petition the government for political change, while at the same time allowing terrorists the freedom of movement necessary to conduct an effective, violent campaign.Footnote 3 The inherent duality of the nature of the democracy–terrorism nexus – that democratic structures encourage legal political action while limiting the state's capacity to crack down on illegal political action – has resulted in a growing scholarly effort to engage in more complex theorizing.
One particularly interesting avenue of research addresses the potential for variation in terrorist violence across democratic states. Perhaps the best developed of these recent efforts conceives of changes in terrorist targeting as being contingent on variations in the ‘openness’ of democratic structures. Several scholars argue that democratic states with institutions which increase minority representation in government are less likely to experience armed rebellion or terrorism than those with less inclusive structures.Footnote 4 This argument rests on Duverger's claim that polities with majoritarian-style electoral rules produce two-party systems with a circumscribed representation of political views, leading to the marginalization of some minority positions.Footnote 5 Without recourse to legal means by which to change policy, these marginalized actors are more likely to turn to violence. Conversely, more inclusive (typically, proportional) electoral systems ‘produce legislatures that often represent the preferences of all citizens’,Footnote 6 thereby diminishing the necessity for violent activism.
This article seeks to develop a counterpoint to this ‘representative inclusiveness’ argument. Simply put, there is neither an unambiguous theoretical rationale underlying nor conclusive empirical support for a central assumption of the inclusiveness argument: that greater proportionality necessarily translates into less terrorism. We build upon important literature in comparative politics which questions the use of static institutional measures of potential inclusiveness as indicators of multipartism,Footnote 7 highlights the ill-effects of fractionalization on regime stability in proportional systems,Footnote 8 and stresses the disconnect between electoral and policy outcomes in proportional systems.Footnote 9 Following from these insights, we argue that greater proportionality actually increases the likelihood of terrorism when instituted in fractionalized societies. In societies characterized by fractionalization, proportionalism includes numerous smaller political groups in the legislature, engendering ‘extreme pluralism’, which simultaneously increases regime instability and minimizes the legislative influence of extreme parties. In such circumstances, proportionalism institutionalizes extreme groups’ political impotence, increasing the likelihood of their use of violence against the state.
We test the implications of these refinements via cross-national statistical analyses of domestic terrorism in a global analysis of 102 democratic states for the period 1975–2007. Our analyses reveal that although majoritarian regimes have been shown elsewhere to experience elevated levels of transnational terrorism, the use of plurality (first-past-the-post) voting rules is not independently related to domestic terrorism in our global sample. We also demonstrate that increases in the national average of members per district, a key indicator of the degree of proportionalism, have significant (though somewhat mild) exacerbating influences on domestic terrorism levels. Finally, and most importantly, we show that the combination of increasing institutional proportionality and increasing social heterogeneity within society increases terrorism, in direct refutation of the logic of previous scholarship and in accordance with our expectations.
Terrorist Strategy, Democracy and Electoral Systems
In general, scholarly work has revealed a complex relationship between regime openness and the likelihood of terrorist violence: put simply, there are about as many reasons to believe that democracy increases the capacity (and, to a lesser extent, the willingness) of disaffected political actors to engage in terrorism as there are reasons to believe that democracy reduces the incentives for political actors to engage in terrorism. In the aggregate, the former view posits that democratic governance is more ‘permissive’ of terrorist campaigns. The decentralized government decision-making structure of democratic polities limits their capacity to develop and support robust and unified counterterrorist strategies.Footnote 10 Moreover, the general guarantee of civil rights and liberties in democracies allows terrorists greater recruiting, propaganda and operational freedom.Footnote 11 Furthermore, in so far as their counterterrorist responses involve the use of force and/or the limitation of civil liberties, democratic governments run the risk of delegitimizing themselves. In extreme circumstances, harsh responses can have the counterproductive effect of swelling the ranks of the insurgency,Footnote 12 and the unpopularity of such responses amongst the voting public can conceivably threaten the tenure of the government.Footnote 13 Alternatively, it is possible that democratic institutions reduce the likelihood of terrorist violence.Footnote 14 Democracy increases the number of non-violent avenues for political actors to redress their grievances;Footnote 15 moreover, democratic participation and elections increase the general satisfaction and political efficacy of citizens, subsequently reducing grievances, thwarting terrorist recruitment and raising the public's tolerance for counterterrorist policies.Footnote 16
Of particular interest in recent research is the degree to which democratic institutions increase political efficacy. Democratic institutions vary widely in the extent to which political interests are translated into legislative representation, and greater representation is likely to decrease functional disenfranchisement and increase political efficacy. Thus, citizens in democratic states with more inclusive political institutions may have fewer incentives to engage in violent coercive behaviour than citizens in democratic states with less inclusive institutions. Though not derived in specific relation to terrorism, Reynal-Querol's game-theoretic treatment of the influence of democratic system type on civil war deduces that proportional systems hold a distinct advantage in regards to systemic stability. In proportional systems, where the legislative representation of more politically extreme groups is necessarily greater than it is in majoritarian systems, the policies ultimately implemented are more likely to approximate the mean of all political groups’ preferred policies. As such, it is more likely that all groups in a proportional system will prefer to support the implemented policies as opposed to rebellion. Put differently, proportionalism systematically increases the opportunity costs of rebellion for any given political group by decreasing the likelihood that the ‘distance’ between any group's preferred policy and the policy ultimately implemented is ‘greater’ than the material costs and risks of rebellion. Reynal-Querol's empirical analyses indicate that proportionalism does decrease the likelihood of civil war relative to presidential, majoritarian and authoritarian systems.Footnote 17
This logic has been adapted directly to the study of terrorism. Quan Li, generally uncovering evidence of both dampening and enhancing effects of democracy on terrorism, finds that democratic systems that minimize avenues for direct political representation are the target of larger numbers of transnational terrorist attacks. Specifically, his analyses show that ‘the proportional system experiences fewer terrorist incidents than either the majoritarian or mixed system’, indicating that greater proportionality translates into greater inclusiveness and less transnational terrorism.Footnote 18 A similar conclusion is reached by Deniz Aksoy and David Carter in respect to levels of domestic terrorism and the emergence of new domestic terrorist groups. Effective disenfranchisement should have a more regular effect on domestic terrorism than transnational terrorism, since the process by which institutional inclusiveness increases efficacy is most germane to domestic interests and actors. Their study provides some evidence that fewer domestic-originating attacks occur and fewer new domestic groups emerge in democratic states in Western Europe characterized by higher levels of structural inclusiveness.Footnote 19
A Reformulation of the Inclusiveness–Terrorism Linkage
While compelling, we believe the ‘representative permissiveness’ account to be problematic in two important ways. Our first criticism involves the use of institutional measures as proxies for the actual inclusion of varied political interests into the legislature. While such measures reflect an institutional capacity to translate multiple political interests into multiple legislative parties, they give no indication of the ex ante pressure to translate more interests into more parties. This is because such measures do not account for the presence or absence of multiple existing political interests in a given society. Crucially, social cleavages interact with electoral systems to determine the establishment of political parties:
Electoral institutions modify the effect of social forces on the creation of political parties. Social forces create more or less pressure for the multiplication of political parties and electoral laws either permit these pressures to be realized or they constrain them by discouraging the formation of new parties … Consequently, unconditional comparative static predictions about the effect of electoral laws on the number of parties are necessarily indeterminate (emphasis added).Footnote 20
As such, the exclusive use of institutional measures, by lending identical explanatory weight to those states which have a pressing ‘need’ (in terms of ensuring broader inclusion) to form new parties and to those which do not, potentially muddles empirical analysis. For example, countries with few existing socio-political cleavages are relatively unlikely to experience domestic political violence regardless of whether or not their institutions are relatively more inclusive, simply because limited inclusion may not disenfranchise any groups. By the same token, countries with numerous socio-political cleavages and moderate institutional inclusiveness should still be expected to face greater risks of violence than states in the former example. In this scenario, the necessary empirical assumption embodied by the exclusive use of institutional measures – that the latter type of state should experience less terrorism than the former simply because its institutions are more inclusive – is clearly antithetical to the logic of the permissiveness account. In all, we believe that tests of this and associated arguments would be much improved by gauging the commensurability of institutional inclusiveness and societal pressures for inclusion.
Moreover, work in comparative politics would suggest that the legislative result of the coincidence of permissive institutions and multiple socio-political cleavages may actually increase political violence. Indeed, our primary substantive critique of the permissiveness argument is that greater political representation simply may not increase political efficacy to a degree sufficient to dissuade the use of violence by some actors – and, in particular, the use of terrorism by small political groups. It must be reiterated that the theoretical underpinnings of the permissiveness argument are clearly most applicable to relatively large political groups within democracies.Footnote 21 Given increased institutional inclusiveness (such as proportional representation), larger groups will send larger contingents to the legislature, thus giving them a greater independent capacity to influence policy making (or, using Reynal-Querol's terminology, a greater independent capacity to pull the mean of all groups’ preferred policy closer to their ideal points).
But when institutional structures include relatively smaller groups in the political process, the distance between their preferred policy points and the mean of all groups’ preferred policy points is not necessarily decreased. Most obviously, the legislative ‘clout’ of any given smaller group is minimal, meaning that such parties lack the independent influence over the legislative mean policy point that inclusion affords their numerically larger counterparts. Thus, by extension, the likelihood that inclusion places the mean policy point within a given small group's range of acceptable outcomes is largely dependent upon the preferred policy points of other included parties. As such, the aggregate effect of greater institutional inclusiveness on the relationship between the mean legislative policy point and the preferred policy points of small political groups would seem to be indeterminate, all else being equal.
Majoritarian rules for the passage of legislation in proportional systems can cause some voters who prefer more extreme small parties to support more moderate coalition parties in an attempt to ‘edge policy closer to their most preferred outcome’. Adherents to extreme parties thus often view votes for their own parties as wasteful, given the lack of influence such parties will wield over policy outcomes. Votes for extreme small parties do occur, but they are normally made (a) on relatively rare occasions by putative moderates who support parties more extreme than their ideal point to pull national policy closer to that ideal; and (b) with greater regularity by ‘true believers’, or those whose commitment to a party's platform trumps their interest in strategic voting.Footnote 22
Based on these considerations, we posit that the inclusion of multiple small groups into the political space increases the likelihood of terrorist violence by raising the likelihood that extreme groups are rendered impotent in the national political process. The notion of ‘extreme pluralism’ is useful in illustrating the nuances of the process by which the marginalization of small political groups can take place – even when those groups accept inclusion in principle at the outset. The presence of several small parties in the legislature can induce volatility in coalition formation, thus leading (especially in cases of pronounced societal polarization) to legislative gridlock and increased fractionalization in proportional systems. Indeed, the coincidence of many parties and significant ideological distances between those parties is potentially threatening to regime stability in several direct and indirect ways. First, fractionalized multipartism increases the likelihood that votes flow away from the centre and towards the extremes. However, given ideological diversity, it is unlikely that these extreme parties can mount a unified challenge to any centrist government. Complicating matters, since fractionalized multipartism compels parties to distinguish themselves from one another, politics becomes more ideological, resulting in doctrinaire platforms and memberships and further decreasing the likelihood of coalition-building and compromise. Finally, since such parties are unlikely to become the linchpin of any government, they have incentives to make grandiose promises and severe demands on the government (which are likely to resonate with those extreme parties’ ‘true believer’ constituencies). As these promises and demands escalate, challenges to governmental efficacy can arise. In all, these centrifugal forces can render impossible the integration of diverse parties into the system, thus inviting constitutional irresponsibility on the part of the government and the employment of extra-constitutional means of protest (such as political violence) by those parties.Footnote 23
By this accounting, even when it leads to the formation of nominally centrist coalitions, proportionalism in fractionalized societies can simultaneously include ‘extreme’ groups in the political process and fail to bring the mean legislative policy point within their ranges of policy acceptability. In such circumstances, inclusion through proportionalism frustrates extreme groups by institutionalizing their political ineffectiveness. This then has the effect of increasing the likelihood that they remain or become comparatively small and weak anti-system or protest movements, which one would speculate are particularly given to the use of terrorism as a tactic.
A potential illustration of this phenomenon involves the now-banned Basque separatist party Batasuna in Spain. Batasuna represented the Basque region of northern Spain from the post-Franco period until it was banned in 2003. Throughout this period it is recognized as having operated as the legal political wing of ETA. Its members participated in multiple national and local elections and reached a peak by winning five seats in the Spanish national parliament in 1986, but their representatives rarely attended sessions. Indeed, their frustration at the impotence of their role in parliament culminated with their boycott of the 2000 election.Footnote 24 Despite being able to win seats to the national parliament (in a polity with a lower legislative tier governed by proportional representation rules), Batasuna never saw its policy preferences accommodated by the range of legislation produced therein, and apparently opted instead to sanction the pursuit of Basque independence through terrorist violence.
By the same token, any significant within-group variability regarding the range of an ‘extreme’ group's policy acceptability may increase the likelihood of political violence. Assume, for example, any scenario in which there exists a division between elements of an extreme group on the basis of the willingness to accept limited policy change (with moderates willing and radicals unwilling to accept limited changes). In so far as inclusion results in any increase in the probability that group policy demands will be satisfied to any degree via the national political process, moderates are likely to accept inclusion. The likelihood that radicals will accept inclusion, by contrast, varies in proportion with the likelihood that inclusion will satisfy their demands for more comprehensive policy change. If inclusion fundamentally shifts the legislative balance of power in such a way as to empower the group with at least something of a direct say over the course of national policy, the likelihood that radical demands for comprehensive change will be satisfied increases. If, however, inclusion does not meaningfully affect the legislative balance of power – as is normally the case with relatively small groups – then inclusion is, in the eyes of radicals, tantamount to the political institutionalization of the movement's marginal national status, and moderates’ acceptance of inclusion may result in the defection of the radicals. Since ‘legitimate’ (i.e., within-system) political avenues for acceptable policy change have been exhausted, radicals are more likely to adopt violent stances than to accept even more pronounced minority status as an even smaller included party.
A useful illustration of this ‘splitting’ behaviour over the willingness to accept national legislative inclusion involves the defection of the Naxalites from the Communist Party of India (Marxist) (or CPI-M) in 1968. In the run-up to the 1967 Lok Sabha elections, the Central Committee of the CPI-M (which had itself split from the Communist Party of India in 1964 over ideological issues) decided to pursue alliances with ideologically dissimilar parties in West Bengal to increase efficacy in national policy making.Footnote 25 This decision was roundly criticized by both communist governments around the world and CPI-M hardliners as an abandonment of ideological principals. The hardline uprising caused the CPI-M leadership to commit to courting alliances with only leftist parties. However, after the CPI-M won 4 per cent of the seats in the 1967 election and the violent repression of a labour uprising in Naxalbari, West Bengal (where the provincial government were dominated by CPI-M deputies), the hardliners rejected political avenues and defected, establishing a violent revolutionary movement known as the Naxalites.Footnote 26 Naxalite terrorism persists to this day in West Bengal and other areas of India.
Thus, in sum, we arrive at two hypotheses regarding the relationships between electoral rules, small-party representation in the legislature, and domestic terrorism. First, from an institutional perspective, we posit that societal fractionalization is a crucial intervening variable in the relationship between institutional inclusiveness and domestic terrorism. In particular, and contrary to the expectations of the permissiveness argument, the proportional electoral rules in a fractionalized society may institutionalize the political weakness of and exacerbate policy preference divisions within extreme groups. Conversely, the proportional inclusion of fewer, larger groups in non-fractionalized societies is generally likely to diminish terrorism, since the relative strength of these groups offers them a meaningful say in the direction of national policy and encourages radical elements to choose intraparty strategies to achieve political influence over defection. Secondly, from a legislative perspective, the foregoing logic would suggest that the mere presence of numerous smaller parties in the legislature would be associated with increased domestic terrorism.
HYPOTHESIS 1: Increasing institutional inclusiveness increases the likelihood of domestic terrorism in democratic states with high levels of societal fractionalization.
HYPOTHESIS 2: Domestic terrorism in a democratic state increases as the number of small parties represented in that state's legislature increases.
Research Design
Our empirical tests of these hypotheses examine domestic terrorism in all democraciesFootnote 27 from 1975 to 2007, and we use the state-year as the unit of analysis (e.g., Botswana, 1985; Colombia, 1990). To construct the dependent variable, we operationalize the annual count of domestic terrorist events.Footnote 28 The general source is the Global Terrorism Database (GTD), which defines a terrorist attack as an ‘intentional act of violence or threat of violence by a non-state actor’ – to include assassinations, hostage-taking, armed and unarmed assaults, bombings, hijackings and facility or infrastructure attacks – and which further meets two of the following three qualifying criteria: (a) is ‘aimed at attaining a political, economic, religious, or social goal’; (b) included evidence of an intention to ‘coerce, intimidate, or convey some other message to a larger audience (or audiences) other than the immediate victims’; and (c) is ‘outside the context of legitimate wartime activities’.Footnote 29 Recent research has disaggregated the GTD data into domestic and transnational terrorism, with attacks in the former category being those in which the ‘venue, target, and perpetrators are all from the same country’ (for example, by the ETA against Spanish citizens in Spain; by Action Directe against French governmental offices in Paris).Footnote 30 As noted previously, such attacks are significantly better suited than transnational events to test hypotheses about the influence of domestic institutional inclusiveness on the decision of domestic actors to resort to political violence. Moreover, the broad spatial coverage of the GTD data allows for a global assessment of democratic institutional and socio-political variation, thus representing an improvement upon analyses employing regional compilations of domestic terrorism.Footnote 31 The GTD data includes 26,877 domestic events for the period 1975–2007. Importantly, the data also code failed and foiled attempted terrorist attacks, the inclusion of which allows us to avoid systematically omitting behaviour on the basis of its success.Footnote 32
Given that the dependent variable takes the form of a zero or a positive integer, event count modelling is more appropriate than ordinary least squares regression. Since ‘over-dispersion’ appears in the data series (as evidenced by standard deviations that are larger than means), we employ negative binomial models.Footnote 33 To address temporal dependence in the dependent variable – which, as indicated by diagnostic tests, follows a first-order trend – we estimate AR(1) generalized estimating equation negative binomial models, which control for first-order autoregression in the idiosyncratic error term of panel data when numerous panels are analysed and when some panels are invariant (i.e., when terrorism equals zero in all years for some countries).Footnote 34
To examine the effects of political inclusion on terrorism comprehensively, we first run statistical models utilizing measures of both the institutional facilitation of political inclusiveness and the actual degree to which societal groups are included into the legislative political process. Two institutional variables, both culled from the 2010 edition of the Database of Political Institutions,Footnote 35 are employed: (a) a dichotomous measure indicating whether or not a democracy employs first-past-the-post (i.e., plurality) in any of its national voting rules; and (b) the average national mean number of representatives elected per district in all directly elected houses of parliament (or overall district magnitude). While the use of a simple ‘majoritarianism’ measure allows for comparability to previous research on the institutional correlates of terrorism, the inclusion of overall district magnitude allows for more nuanced tests of the gradient effects of inclusiveness. Indeed, much comparative politics scholarship identifies district magnitude as being central to the representative inclusiveness of a given system (with greater district magnitude predicting greater inclusiveness), irrespective of the number of political parties produced by the system.Footnote 36 It has been suggested that ‘as district magnitude increases, each party's share of seats tend to correspond more closely to its vote share.’Footnote 37 In the main, the ‘representative permissiveness’ logic would expect each of these measures to be negatively and significantly associated with domestic terrorism; conversely, the argument developed here does not expect either of these static measures of institutional permissiveness to be significantly related to domestic terrorism.
In addition to these institutional measures, and in line with our critiques of the permissiveness argument, we include one measure of the commensurability of socio-political cleavage and institutional permissiveness, and one measure of the degree to which small parties are present in the legislature. First, in order to use ethnic diversity as a proxy for gauging the multiplicity of social forces, we multiplicatively interact the variable Ethnolingusitic Fractionalization (or ELF) with our two measures of institutional inclusiveness. ELF, a percentage measure indicating the likelihood that any two randomly selected citizens of a given country are from different ethno-lingusitic groups, is a widely utilized measure of the diversity of societal groups.Footnote 38 In general, a positive association of ELF x Majoritarian and a negative association of ELF × Overall District Magnitude with terrorism would provide support for the permissiveness argument. Conversely, a negative association of ELF × Majoritarian and a positive association of ELF × Overall District Magnitude with terrorism would provide support for Hypothesis 1. Secondly, we include in separate models a measure of the Legislative Fractionalization of the legislature, or the likelihood that any two legislators drawn at random are members of different political parties. As this variable is widely used as a measure of the number of small parties in a given system,Footnote 39 a positive and significant association of this variable with domestic terrorism would provide support for Hypothesis 2.
We also include several control variables in our analyses. First, we control for country-specific heterogeneity by including a measure of the demographic size of the country (Natural Log of Population). Much of the variation in the timing of terrorist attacks can be accounted for through variation in country size.Footnote 40 Secondly, we include a measure of a state's military capabilities to control for the state's capacity to combat terrorism. Military Capability, operationalized as the annual composite percentage index of a state's share of the world's total material capabilities, is taken from the Correlates of War (COW) Project's Composite Index of National Capabilities,Footnote 41 and is expected to be negatively related to the dependent variable. Thirdly, to account for the influence of material wealth on terrorist proclivities, we include for each country-year the Natural Log of Per Capita Gross Domestic Product, taken from the Penn World Tables.Footnote 42 We conceive of this variable as capturing general satisfaction with the economic status quo, which might both dampen the importance of representative inclusiveness to citizens and decrease their grievances against the government.
Fourthly, to control for regime stability, we specify Continuous Democracy Years, culled from the POLITY 4e dataset. Previous literature has argued and found that unstable or nascent regimes experience greater levels of terrorism than stable regimes.Footnote 43 As such, we expect a negative relationship between this variable and terrorism. Fifthly, to account for the potential capacity of an independent executive to encroach on civil liberties (thus perhaps increasing the prospect of government deadlock, terrorist incentives to elicit a harsh overreaction by the government, and ultimately weakening the capacity of the state to fight terrorism),Footnote 44 we include a dichotomous measure of presidentialism, with the expectation of a positive influence on the dependent variable.Footnote 45 Sixthly, it is possible that the restructuring of the international system after the US–Soviet Cold War rivalry diminished the strategic costs to states of granting nationalist groups autonomy or independence, thus diminishing the need for those groups to engage in violence.Footnote 46 We therefore include the dichotomous variable Post Cold War, coded 1 for country-years after 1991 and 0 otherwise. Finally, as involvement in international conflicts could sap states’ capacities to conduct effective domestic counterterrorism, we include a dichotomous measure of International War Involvement, as coded by the COW dataset.Footnote 47
Results
The results of the various analyses using generalized estimating equations (GEE) are presented in Table 1. First, as predicted by Clark and Golder but broadly contrary to the expectations of previous terrorism scholarship,Footnote 48 the unconditional variable Majoritarian System (Model 1) is not significantly associated with domestic terrorism.Footnote 49 The unconditional measure of overall district magnitude, by contrast, is positively and significantly related to domestic terrorism. While unreported marginal effects analyses indicate that this effect is substantively quite small, and the unconditional effect of the variable is rendered statistically insignificant when extreme outliers are removed from the analysis, it is worth reiterating that the relationship is not in line with the expectations of the ‘institutional permissiveness’ framework. In terms of the control variables, Natural Log of Population, Years of Continuous Democracy and Presidential System are consistently and significantly related to the dependent variable in the expected manner.Footnote 50
Table 1 Inclusiveness, Fractionalization and Domestic Terrorism in Democracies, 1975–2007 (Negative Binomial GEE Models)
Note: First degree autocorrelation AR1 specified. Z-Scores in parentheses. *p < 0.05; **p < 0.01 (one-tailed tests).
Secondly, as predicted in Hypothesis 1 and contrary to the expectations of previous scholarship, the coefficient for ELF × Majoritarianism in Model 3 is negative and significant, which indicates that terrorism is more likely in non-majoritarian systems when fractionalization is high. Moreover, the coefficient for ELF × Overall District Magnitude in Model 4 is positive and significant, showing that terrorism is more likely as both fractionalization and district-based proportionalism increase. Marginal effects analyses (using a modified ‘bootstrapping’ method for multiplicative interaction terms, and holding all remaining variables at their mean – or 0 if dichotomous) can help shed additional light on these base findings.Footnote 51 In regards to ELF × Majoritarianism, the marginal effects indicate that the statistical significance of the GEE model is in fact relevant to non-majoritarian systems: while the negative effects of increasing fractionalization on terrorism given majoritarianism are statistically insignificant, the positive effects of increasing fractionalization on terrorism given non-majoritarianism are significant. Figure 1 tracks this latter trend across various levels of fractionalization, with labels indicating the general ELF ‘location’ of several non-majoritarian states.
Fig. 1 Marginal effects of ethno-linguistic fractionalization and domestic terrorism in non-majoritarian democracies, 1975–2007: Dashed lines represent 90 per cent confidence intervals Note: PV [Y|X] when ELF = 0: 2.23 Note: Dashed lines represent 90 per cent confidence intervals
When ELF = 0 (as it does for several non-majoritarian ‘nation-states’ such as Australia, Italy, Portugal and Sweden), the predicted number of terrorist attacks is 2.23. However, as ELF increases, the predicted number of attacks rises, with an average increase of 7 predicted attacks per 10 per cent increase in ELF. This increase is greater at the highest levels of ELF, but there is little indication that this is more attributable to outliers than to the general upward trend; indeed, the average annual number of attacks of the high-end ELF outliers (Israel and South Africa) is one-fifth that of those states whose ELF scores are within one standard deviation of the population mean of ELF (to include such ‘high-terrorism’ states as Peru, Sri Lanka, Ecuador and Colombia). In any event, the findings of all interaction models are robust to the exclusion of the high-end ELF outliers.
Figure 2 is a panel comprised of several analyses of the conditional effects of ELF and Overall District Magnitude on domestic terrorism. These analyses broadly reveal that the effects of increasing ELF on terrorism are statistically insignificant for a considerable portion of that variable's distribution. However, beginning with a district magnitude of 28,Footnote 52 the relationship between increasing ELF and domestic terrorism is positive and statistically significant. Therefore, and quite ironically given the arguments of previous scholarship, it would appear that increases in institutional inclusiveness are associated with relatively lower levels of terrorism when there exists little or no social diversity (at least as measured by ELF). Moreover, the relatively high degree of proportionalism embodied by district magnitudes at this level (comprising only 8 per cent of all observations) might suggest that ‘extreme pluralism’ in such societies empowers a plethora of small parties, which introduces legislative gridlock and makes violence more likely.Footnote 53 At the same time, it must be noted that the exclusion of extremely high district magnitude outliers (including Russia, Israel and the Netherlands) does not alter the substance or statistical significance of these findings. Overall, the analyses provide strong support for Hypothesis 1.
Fig. 2 Marginal effects of increasing ethno-linguistic fractionalization on domestic terrorism at various levels of overall district magnitude, 1975–2007
Thirdly, the coefficient for Legislative Fractionalization (Model 2) is positive and significant, indicating that the greater the number of small parties, the greater the likelihood of domestic terrorism. Figure 3 tracks the marginal effects of increasing fractionalization on terrorism and, though the substantive effect is rather small, it is statistically significant at all levels and supportive of Hypothesis 2. Moreover, the statistical effect remains significant despite the removal of high-end outliers, and the coefficients for quadratic forms of the variable are insignificant, indicating clear linearity in the relationship between the number of small parties and terrorism.
Fig. 3 Marginal effects of legislative party fractionalization and domestic terrorism in democracies, 1975–2007
Finally, several additional statistical tests broadly confirm the robustness of the base findings. Analyses which include terrorist attacks in contentious territories of democratic countries (such as the West Bank and Gaza for Israel, Puerto Rico for the United States) as coded by GTD return results that are substantively identical to those produced using the chosen dependent variable. So, too, do analyses which (a) substitute the number of individuals killed in domestic terrorist attacks per country-year as the dependent variable (and thus account for the intensity of domestic terrorist violence); (b) employ different measures of ethno-social diversity (including the number of overall ethno-politically relevant groups); and (c) employ the generally more stringent variable of mean district magnitude in the lower house of parliament as a measure of institutional permissiveness. Also, the findings regarding ELF × Majoritarianism are robust to the specification of non-population averaged models (i.e., fixed and random effects). Also, the inclusion of a threshold variable (indicating the percentage of seats that need to be won by a party in order to gain representation in the legislature) not only fails to change the nature and significance of the key findings, but is also negatively and significantly related to terrorism. This negative relationship might be viewed as further supporting the findings presented here, in that the inclusion of very small parties into the legislature may intensify the frustration felt by terrorist radicals when their moderate counterparts are ‘co-opted’ as perpetually weak actors in the legislature.
Discussion and Conclusion
Scholars have naturally focused on the role that political institutions play in the generation of violent political events. Domestic political institutions act as an important moderator of both demands and how they are expressed. Previous literature has argued that in terms of terrorism, a permissive environment creates alternative pathways for the expression of grievances, which implies that violence would be minimized in societies with more representative institutions. This logic is based on two flawed assumptions. First, it assumes that extreme groups view the political systemas legitimate and want to be co-opted by it. Secondly, there is the implicit assumption that a small, marginalized extreme group would be able actually to move policy outcomes in a meaningful way. We believe that these assumptions rarely reflect political reality.
In general, groups inclined towards domestic terrorism are likely to feel, at the very least, that the political system is illegitimate. In the most extreme cases, these groups want to tear down the current political system and replace it with something closer to their ideal point. As such, inclusion in the system seems unlikely to assuage their grievances and attempts to do so may further serve to exacerbate the split. In addition, even if these groups attempt to act within the political system, it may only serve to institutionalize their impotence, as they are unlikely to be able to alter policy outcomes in a meaningful manner. The net effect, then, is that openness in and of itself will not decrease domestic terrorism; indeed, when combined with a highly fractionalized society, openness should actually lead to more terrorism.
Our statistical analysis of democratic states and domestic terrorism from 1975 to 2007 confirms our expectations in two critical ways. First, we find little evidence that institutional inclusiveness in and of itself is correlated with domestic terrorism. While this differs from previous research, we believe our models are both more complete and properly focus on domestic terrorism. Secondly, the models demonstrate that high levels of representation, given high degrees of fractionalization, actually increase the amount of domestic terrorism. These situations, where the weakness and inability of these extreme groups becomes both acute and most noticeable, are likely to drive them further into alternative methods of contention, such as terrorism. We discussed this as the process potentially motivating Batasuna's boycott of the national elections in Spain in 2000.
In general, our analysis demonstrates that permissive political systems are not a panacea for limited domestic terrorism. In fact, under some circumstances the openness of the system forces the extremists further from the centre and increases the amount of domestic terrorism. Future research needs to delve more deeply into the process by which permissiveness creates divisiveness. In particular, does this process affect all groups equally? In other words, can some groups be co-opted through a permissive environment even though we have seen that the average effect is to drive them towards violence? Regardless, it is clear that our analysis has revealed that the effect of democratic institutions is neither obvious nor simple, and that a more nuanced approach is needed to determine the relevant relationships.
