The political economy of freedom, democracy and transnational ...

3 downloads 0 Views 442KB Size Report
Jul 21, 2006 - Cite this article as: Kurrild-Klitgaard, P., Justesen, M.K. & Klemmensen, R. Public Choice (2006) 128: 289. doi:10.1007/s11127-006-9055-7.
Public Choice (2006) 128:289–315 DOI 10.1007/s11127-006-9055-7 ORIGINAL ARTICLE

The political economy of freedom, democracy and transnational terrorism∗ Peter Kurrild-Klitgaard · Mogens K. Justesen · Robert Klemmensen

 C

Springer Science + Business Media B.V. 2006

Abstract We conduct an empirical analysis of data relating measures of economic and political freedom to the occurrence of transnational terrorism 1996–2002. We use binary logistical regression models to predict the probablities that a country will experience transnational terrorist attacks and that a given terrorist originates in a particular country. We find that the extent of political rights and civil liberties is negatively related with the generation of transnational terrorists from a country, but where the former is also negatively related with the occurrence of transnational terrorism in a country, the latter exhibits a non-linear relationship. A number of alternative explanations are disconfirmed: transnational terrorism is unrelated to inequality, economic growth, education, poverty, etc., while a society’s fractionalization has mixed importance, and the religious composition has no or little association with attracting or producing transnational terrorism. A more trade-oriented economy seems consistently to associate with smaller probabilities of a country experiencing and generating transnational terrorism. Keywords Bush doctrine . Democracy . Freedom . Rent-seeking . Transnational terrorism We are led, by events and common sense, to one conclusion: The survival of liberty in our land increasingly depends on the success of liberty in other lands. The best hope for peace in our world is the expansion of freedom in all the world. George W. Bush, 2nd Inaugural Address, 20 January 2005 ∗ An earlier version was presented at the Public Choice conference on The Political Economy of Terrorism, George Mason University, 24–25 May 2005, and we owe thanks for comments to the participants in that event.

P. Kurrild-Klitgaard Dept. of Political Science, University of Copenhagen, DK-1014 Copenhagen, Denmark e-mail: [email protected] M. K. Justesen · R. Klemmensen Dept. of Political Science and Public Management, University of Southern Denmark, Campusvej 55, DK-5230 Odense M, Denmark e-mails: [email protected]; [email protected] Springer

290

Public Choice (2006) 128:289–315

1. Introduction Do terrorists act more like fanatical zealots moved by concerns of justice or more like rational actors pursuing private benefits? And does the character of the political-economic institutional arrangements of a country have any impact on the occurrence of terrorist acts against it? Or the production of terrorists who attack other countries? Such questions have become dramatically relevant in recent years. In the aftermath of 9/11 and all the havoc caused in the Middle East (and elsewhere) by Islamist terror groups (and others), it would be difficult not to see religion as a powerful force in the explanation. Yet other explanations have focused on the seemingly increased levels of anti-Western terrorism as the result of, e.g., unequal distributions of wealth or simply due to a particularly bellicose and anti-modern religion. In contrast, the view taken by the Bush administration, as well as some academics and observers, is that ultimately the key to a world less marked by conflict – including terrorism – is one where regimes are characterized by individual liberty, free markets, democracy and the rule of law. As such the Bush administration’s doctrine is tied to a normative recommendation but it is one which may be seen as essentially based in a positive institutional analysis, i.e., an analysis proposing that institutional arrangements affect the decisions individuals make (cf., e.g., Ostrom, 1986). Specifically, it is argued that more “freedom” (or “liberty”) will lead to less terrorism and therefore that creating more freedom will lower the amounts of terrorism.1 This reasoning has even gained support of an uninvited kind, namely in a number of Islamist terrorists declaring freedom and democracy as their stated targets.2 Now, can rational choice analysis contribute something to this issue? That is, are there any theoretical reasons why it should be the case that more freedom leads to less terror? And is there any empirical evidence to support such reasoning? The following analysis will try to consider these questions. We will first consider some alternative rational choice analyses looking at the possible relationships between the extent of freedom and terrorism and sketch a set of testable propositions (Section 2). We will then consider various types of relevant data, including variables enabling us to make an examination of alternative explanations (Section 3), and submit the hypotheses to tests (Section 4). We will then interpret our results and compare them with some other recent studies (Section 5).

2. Rational choice aspects of freedom, democracy and terrorism The term “terrorism” is one of those controversial and essentially contested concepts within the discourse of politics and accordingly also in social science research. For the present purposes we may first and foremost note that irrespective of how terrorism is defined, it must be seen as merely a sub-branch of “political violence” more broadly defined, i.e., the deliberate use of violence by individuals or groups as a means with which to achieve political goals (no matter the more specific contents of these). As such terrorism – as a form of political violence – is related to such other phenomena as coups, wars, civil wars, genocide, etc., as well as rebellion more broadly defined (cf., e.g., Kotowski, 1984; Finney, 1987;

1 For statements of this line of reasoning see, e.g., Bush’s “Forward Strategy of Freedom” (announced 6 November 2003), his 2nd Inaugural Address (20 January 2005), and his 2005 State of the Union speech (5 February 2005), and in particular his London speech (19 November 2003). 2

Cf., e.g., Abu Musab al-Zarqawi’s message prior to the first free Iraqi elections, recording, 23 January 2005. Springer

Public Choice (2006) 128:289–315

291

Kurrild-Klitgaard, 1997). Terrorism may then more narrowly be seen as the deliberate use of violence against non-combatants for political purposes.3 How may rational choice theorists approach the study of terrorism and other forms of political violence? The obvious point is, of course, to describe the behavior of those engaging in such activities as a deliberate choice, involving costs, benefits and externalities, as well as focusing on the strategic interaction with other actors, not least the governments or other political opponents against which such acts may be directed (Lichbach, 1995). Such an approach has been gaining ground in recent decades, yet most scholars who have investigated the larger “causes” that may influence terrorist behavior have looked at the ideologies of the perpetrators or their social and cultural circumstances; it should nonetheless be obvious that the types of political-economic institutional arrangements underlying the structure of the societies where political violence take place might also influence the rational behavior of both governments and dissidents alike – even scholars have so far not quite reached agreement on exactly what variables are important, or how they may encourage or discourage political violence (cf. Lichbach, 1987). In the following we shall briefly present and survey some of those theories formulated in recent decades about the potential relationships between on the one hand economic and/or political freedom and on the other hand terrorism. 2.1. Some theoretical models In an early application of rational choice theory to the study of terrorism Richard M. Kirk suggested that terrorists should be seen as rational actors, but that they should not be seen as pursuing what they hold to be public goods, since this would necessarily entail freerider problems (Kirk, 1983: 42f; cf. Tullock, 1971, 1974; Kurrild-Klitgaard, 2004). Rather, governments supply a number of private benefits, and terrorists should be seen as interest groups pursuing private benefits, i.e. as rent-seekers (cf. Tullock, [1967] 2004; Buchanan, 1980; Tullock, 2005). Kirk’s reasoning is this: [The] operation of government results in the generation of financial residuals or rents. It is the existence of these rents that is the motivating force behind acts of political terrorism. Thus terrorist organizations, in an effort to capture some of these rents, will act so as to maximize the profits from their violent activities. . . . Terrorist activity is considered to be a threat to the government’s monopoly power over determination of the distribution of rents. . . . [When] the cost of gaining conventional political influence is high enough, or some other explicit or implicit barrier to entry into the political sphere exists, the use of violence in the form of political terrorism can become a profitable method of rent seeking. (Kirk, 1983: 43f) In his analysis Kirk formulates a simple model of a process of rent-generating governments, which seek to avoid social losses while simultaneously “skimming” rents off government activities, and profit maximizing terrorists who seek to capture these rents through acts, which are either violent or non-violent, depending on what is relatively most efficient. Kirk’s analysis concludes that the equilibrium outcome always will be one in which there is some terrorist activity, but also that there may be situations with two equilibria: one with a relatively large government and much rent-seeking by terrorists and one with a relatively 3

Cf. McCormick, who defines it as “the deliberate use of symbolic violence or the threat of violence against non-combatants for political purposes.” (McCormick, 2003: 474). We have found the inclusion of symbolic violence and of threats to be unnecessary and problematic to use for empirical studies. Springer

292

Public Choice (2006) 128:289–315

smaller government with less terrorism (Kirk, 1983: 44–48). While the relationship between the two variables – government size and terrorism – is not necessarily strictly linear, the model does indicate that a smaller, less intrusive government – which will constitute less of a prize to rent-seekers – will tend to associate with lower levels of terrorism, while a larger government should be associated with higher levels of terrorism. Kirk’s analysis is different from but essentially compatible with what has been suggested and investigated by Rudolph J. Rummel. Rummel’s analysis is non-formal but largely compatible with a rational choice analysis: He argues that free people are less likely to engage in violent acts against each other, for the simple reason that interaction between free actors tends to be a plus-sum game, while concentration of power turns human interaction into zero-sum or negative-sum games. Such a society is characterized by personal, economic and political freedom is a multi-dimensional field of diverse social forces – some intersecting, some opposing, some overlapping. The net-effect is to cross-pressure interests, to cross-cut status and classes, and thus inhibit the growth of societywide violence. As a society becomes more authoritarian or coercive, however, the spontaneity of a social field declines, social forces become polarized, the multidimensionality of interests is reduced. Interests and issues begin to revolve around a single dimension: one’s political power. The dividing line between the ‘ins’ and ‘outs’ becomes a conflict front across society along which extreme violence can occur. (Rummel, 1985: 420) Where people are left free to pursue their own desires and the government is held responsible for its actions through elections, a “spontaneous social field” is created within which humans are most secure – violence is minimal, and human and economic development are best achieved. Rummel has formulated this as a specific hypothesis, the “Freedom/Domestic Violence Proposition”: “Freedom inhibits domestic violence (that is, the more libertarian a state, the less internally violent it can and tends to become).” (Rummel, 1985: 421; cf. Rummel, 1984; Rummel, 2001: Appendix). However, it should be noted that Rummel focuses on political violence more generally; this also includes the activities of the government and thereby is different and broader in its scope and application from those that focus on terrorism (e.g. Kirk) and more similar to those that look at political violence broadly (e.g. Finney infra). Yet Quan Li has offered theoretical arguments and a hypothesis making a claim similar to what is implied by Rummel’s, namely that better possibilities for democratic participation will tend to reduce terrorism (Li, 2005: 280f). His argument is that democracy lowers the costs of expressing political preferences and of solving conflicts. Li, on the other hand, also argues that institutional constraints on governments in the form of liberties and other “negative” human rights will increase terrorism because it will give terrorists the freedom to communicate and organize (Li, 2005: 281ff; cf. below). Other theorists, inside and outside the rational choice tradition, have taken a different view of the possible relationship between government size and terrorism. They have focused on the possibility that a more extensive and intrusive government will control more resources which it may devote to the suppression of dissent, including political violence, terrorist activities, etc., and they predict that the power of government will be negatively related to violent conflicts. One example of this line of analysis is Louis Finney’s dissertation on the economics of revolution and political violence, which developed an alternative rational choice model of political violence, inspired by work by public choice classics (Olson, [1965] 1971; Frohlich & Oppenheimer, 1974; cf. DeNardo, 1985), and which he tested against empirical Springer

Public Choice (2006) 128:289–315

293

data. Finney’s theoretical model is based on an assumption of strategic interaction between governments and rebels, and his theoretical conclusion is: [The] level of government spending should be correlated with the level of development of a nation, because the greater the share of the income left over from subsistence levels for residual government spending. Thus, government spending should be negatively related with the amount of domestic political violence. Another interpretation . . . is that the level of government spending is an indication of the power and ability of the regime to expropriate and utilize resources from its citizenry. In this case, if a government increases its ability to collect revenue and spend them, that government has more resources at its disposal to quell, control, and appease its citizenry. This view of government spending suggests that the higher the level of government spending, the lower the level of political violence. (Finney, 1987: 196f) But no matter which of the two interpretations is made, Finney’s model suggests that we should expect government size (primarily measured by spending levels) and political violence to be negatively correlated. Finney’s model deals with political violence broadly rather than terrorism, and it does not deal with government powers other than spending, but at some level his reasoning may be seen as similar to Li’s previously mentioned hypothesis that tighter institutional constraints on governments (e.g. in the form of civil liberties) will lead to increased terrorism (Li, 2005: 281ff). Li argues that by restricting the use of government powers to fight terrorism, freedom of association, speech and movement may make it easier for terrorists to organize and operate, just as freedom of the press may increase coverage of terrorist acts and thereby encourage such. This smacks of a complex relationship between how repressive or how liberal a state is and the extent of dissent. Some years ago Ted Gurr and Mark I. Lichbach suggested that the relationship between how repressive a regime is and the levels of political violence might not at all be linear: more powerful and authoritarian regimes may be better at suppressing dissent, but simultaneously may be more likely to generate dissatisfaction (cf. Gurr, 1970; Lichbach & Gurr, 1981; Lichbach, 1987). This type of analysis has subsequently been extended and tested (e.g., Muller & Weede, 1990), and in a relatively recent theoretical model and empirical application quite similar to Lichbach’s Gareth Davis has used rational choice theory to investigate the relationship between repression, relative deprivation and levels of political violence (Davis, 2004). He models rational actors as interested in both political goods and non-political goods, and he too sees their behavior as one where they may choose between different types of political activities (including the use of violence for political purposes) so that when the costs (including opportunity costs) of political violence fall, or when the benefits of non-violent political activities fall, individuals will devote more resources to violent political activities. Davis derives two possible cases: In the one he works from the presumption that repression by a government may be efficient in reducing traditional, nonviolent political activities but that it often will be inefficient in reducing political violence; increased repression will in fact lead to increased political violence, because it makes nonviolent activities relatively less attractive (Davis, 2004: 7f). In the other model, Davis assumes that repression will be effective in suppressing not only non-violent activities but eventually also political violence. On this basis Davis hypothesizes a curve-linear relationship between the extent of repression and the level of political violence in the form of an inverted U, so that very repressive and very non-repressive societies have low levels of political violence, while the latter peaks for semi-repressive societies (Davis, 2004: 9ff). Springer

294

Public Choice (2006) 128:289–315

We may somewhat simplified illustrate the four possible causal claims made in the aforementioned works as done in Figures. 1–4. 2.2. Testable propositions While the models of Rummel, Finney, Gurr/Lichbach and Davis have been submitted to more or less extensive testing by the authors, this has been done on the basis of empirical data somewhat different from each other and now partly dated (e.g. usually not including post 9/11-data), and Kirk’s model has never been explicitly tested. Furthermore, some of the models have been tested against measures of terrorism and others against political violence Fig. 1 The Kirk model

Fig. 2 The Rummel model

Springer

Public Choice (2006) 128:289–315

295

Fig. 3 The Finney model

Fig. 4 The Gurr/Lichbach/Davis model

more broadly defined, and while several have included measures of democracy only one (Rummel’s) has included measures of economic freedom. As such there would seem to be a lack of studies estimating the possible empirical relationships between various forms of freedom and terrorism, yet the basis of the analysis summarized here we may in fact suggest some hypotheses that conceivably might be tested. The first proposition that we may try to test is the possible relationship between the extent of economic freedom and terrorism. Following the Kirk model, we may reason that the more extensively the state intervenes in the economy, the higher are the potential rents to be reaped – that the higher the rent-value of possessing political power, the more attractive it will be for political groups to invest in the use of violence as a tool with which to realize their Springer

296

Public Choice (2006) 128:289–315

favored policies. In other words, the more extensive the state’s economic interference, the higher the level of domestic terrorism. But if economic freedom and government intervention indeed are opposites, then we may assume that the reverse relationship holds true too, i.e., that the more extensive economic freedom, the less terrorism. However, if the focus is upon border-crossing terrorist acts, it would seem more difficult to apply this logic, or perhaps even meaningfully derive any testable hypotheses about economic freedom and terrorist attacks. Nonetheless, we may formulate the following testable hypothesis: Hypothesis 1. Higher degrees of economic freedom leads to less terrorism The second causal relation we may try to test is the possible relationship between political freedom (including both political rights and civil liberties) and terrorism. Rummel’s analysis, as well as the hypothesis of Li, suggests that if the costs of participating in traditional democratic politics are relatively low compared to rebellious/terrorist activities, there will be less of the latter, i.e., that the higher the level/amount of democracy, the lower the level/amount of terrorist activities and other forms of political violence. It is less clear whether the extent of democracy should have any effect for border-crossing terrorism; it might be argued (e.g., along the lines of Finney), that a “weaker” authority might attract more terrorists. But since these two propositions – that democracy will deter terrorism or that it will stimulate terrorism – are polar opposites, we may for practical purposes reasonably seek to test just one of them: Hypothesis 2. Higher degrees of political freedom lead to less terrorism However, political freedom may both analytically and empirically be seen as composed of at least two distinct elements, namely political rights (e.g., the ability to vote, stand for election, etc.) and civil liberties (e.g. freedom of otrganization, expression, etc.), and there might not necessarily be the same relationship between the two and terrorism. We may accordingly seek to test two separate hypotheses: Hypothesis 3A. More extensive political rights lead to less terrorism and Hypothesis 3B. More extensive civil liberties lead to less terrorism Given these hypotheses, we shall in the following statistical analysis develop equations to estimate the following type of function: Y = f (E, P, C)

(1)

where Y is some measure of terrorism, E is a measure of economic freedom and P is one of political rights, while C is one of civil liberties (and where the two latter may be seen as jointly constituting democracy, D). Specifically, we shall assume that as for the signs it will be the case, that where we are considering all three elements, E, P, C < 0 Springer

Public Choice (2006) 128:289–315

297

while the signs will be as such when we consider the two democracy measures combined: E, D < 0

3. Research design, data and methodology In this section we outline the research design used to test the general hypotheses sketched in the previous parts of the paper. We employ a cross-sectional design for a sample of countries (between 97 and 121) and conduct statistical analyses for the period 1996–2002 as a whole and furthermore divide this period into two sub-periods (1996–1999 and 2000–2002) for which we perform the same tests in order to see, if there are differences in the results covering the years before and around/after the 9/11-attacks and the War of Terrorism declared by the Bush-administration. 3.1. Measuring terrorism: The dependent variables For the purpose of this paper we would ideally want a measure of domestic/national terrorism to use as the dependent variable, i.e., terrorism performed by individuals or groups from country X directed at persons, groups, property or the government or unarmed citizens of country X, and which has consequences for the domestic institutions, policies, property and citizens of country X only (Rosendorff & Sandler, 2005: 172). Instead, we shall have to rely on something less ideal, and we have decided instead to use data from the so-called ITERATE dataset (Mickolous, Sandler, & Murdock, 2003), which is among the most widely used datasets for the study of terrorist activities. The ITERATE dataset contains various measures relating to transnational terrorism, i.e. terrorism which involves and has consequences for two or more countries (Mickolous et al., 2003; Rosendorff & Sandler, 2005: 172, 174–75) (see Appendix A for further references).4 However, by using some variables from the ITERATE dataset we may, in a slightly roundabout way, get around the lack of data for strictly national data and still say something about the possible influence of freedom on terrorism; specifically, we shall use two variables from this dataset: One indicating whether a given country has been the victim of a terrorist attack in a given period, and one focusing on the country of origin of the perpetrators of terrorist attacks. As such, our analyses are conducted in two parts. In the first part we attempt to sketch a profile of the countries that are more (or less) likely to be victims of terrorist attacks, and we do so by focusing on the countries in which terrorists attacks are committed. In the second 4

Another measure of terrorism is the so-called Global Terrorism Index constructed by World Market Research Center (http://www.worldmarketsanalysis.com/), an international risk rating company. This variable is attractive in that it focuses not only on cross-national terrorism but on any kind of terrorism, and its dataset includes data for 186 countries and territories. On the other hand, this measure is less useful, since it is not generally and freely available to researchers and only exists for a single year of observations and seems not to be continued. For that reason we found the dataset (which has been used by Abadie, 2004) to be of little value. There is also an index created by R.J. Rummel and measuring political violence in countries (Rummel, 2001); however this also includes acts of government violence against citizens and only exists for a single year and so has been found of less value for present purposes. In order to test the robustness of the results of the present research a number of the tests have also been made using data from the US State Department’s Patterns of Global Terrorism report, which publishes annual accounts of terror incidents in the world. The data from these reports also only contain transnational terrorism; the results are not reported here but are available from the authors upon request. Springer

298

Public Choice (2006) 128:289–315

part we analyze the characteristics of the countries from where the perpetrators of terrorism originate, i.e. countries that – in a certain sense – can be said to “produce” terrorists; we do so by focusing on the nationality of the participating terrorists and by analyzing whether countries that possess certain institutional characteristics are more likely to produce terrorists than others. 3.1.1. Countries and transnational terrorist attacks A host of measures have been proposed to asses and quantify the degree to which countries are plagued by terrorism (cf., e.g., Frey & Luechinger, 2003; Crain & Crain, 2006). In this paper we measure whether a country has been the victim of terror simply by establishing whether or not a terror incident occurred in a given country during the period we are investigating. As a consequence we do not use standard measures of terrorism such as the number of terrorist incidents, the number of deaths, or the total number of casualties caused by terror. The merit of such measures is that they aim to capture not only the presence of terrorism but also the intensity and severity of terrorist incidents. However, the number of terrorist incidents and the exact number of deaths and casualties caused by terror are often difficult to establish exactly – especially in less-developed countries and non-democracies. Since only some subset of all terrorist incidents are reported by the media and figure in official statistics, this makes both measures of terrorist incidents and measures of casualties vulnerable to bias in the reporting on terrorism (Frey & Luechinger, 2003: 7; Drakos & Gofas, 2005). Secondly, it may be difficult to establish – and often depends on subjective opinions – whether a country experiencing few terror incidents causing many casualties is more plagued by terror than a country experiencing several terror incidents causing fewer casualties. For instance, even though the US by a measure of the number of terrorist incidents was not the most heavily affected country in 2001, by a measure of the number of deaths by terror – and the economic consequences thereof – it clearly was (cf. Shughart, 2002; Rathbone & Rowley, 2002: 8). Finally and perhaps most importantly, it is also obvious that the number of casualties in terrorist accidents often is very random and unrelated to institutional or economic factors; a case in point is the 2001 attack on the World Trade Center, where the number of casualties might have been dramatically increased (or lowered) had any of the two planes hit a little lower (or higher) or a little later (or earlier). On methodological grounds, moreover, using the number of persons killed or injured by terror incidents may cause some potentially severe problems. In linear regression models (e.g. OLS), for instance, one major problem is that outlying cases are likely to affect the parameter estimates disproportionately thereby creating a bias in the results. Clearly, performing OLS regression in such a situation means that, e.g., the 9/11 terror incident on US soil will exercise a disproportionate influence on the overall regression line and the β-estimates. In order to create a measure of terrorism that does not suffer from the above weaknesses, we propose a solution where we divide the world into two types of countries: Those that are plagued by terror and those that are not. This makes this dependent variable dichotomous, assuming the value 1 if a country has experienced a terrorist incident in the period and 0 if has not.5 Admittedly, this is a crude measure of terrorism and it is by no means flawless (e.g. it does not discriminate between the intensity and severity of terrorist attacks), but at the very least it does avoid some of the methodological problems previously outlined. Secondly, since we aim to analyze whether countries suffering from terror posses certain economic,

5

We have here used the “location end” variable from the ITERATE database as dependent variable. Springer

Public Choice (2006) 128:289–315

299

political-economic institutional characteristics, a variable separating terror-plagued countries from non-plagued countries will suffice. 3.1.2. Nationalities and transnational terrorists The problem with many analyses focusing solely on the amount terrorist attacks is that they do not shed any light on the characteristics of the perpetrators of terrorism, nor – in the case of transnational terrorism – on the countries from where the perpetrators originate. Although many papers (e.g. Li, 2005) confine themselves to analyses of terrorist attacks per se, we shall in the second part of our analysis attempt to overcome this problem by focusing not on the countries that are attacked but instead on the countries of origin of the perpetrators of terrorist attacks and the characteristics of these. Hence, in this part of the analysis, we relate economic, political and institutional characteristics of countries to the nationality/origin of the terrorists. In order to create a measure indicating whether a certain country has given rise to terrorists or not, we again choose to employ a simple dichotomous measure based on the nationality of the terrorists.6 For a given country, this variable takes on the value 1 if a person from that particular country has been engaged in terrorist activity in the period we are investigating, and 0 if not. This measure can be viewed as a proxy indicating whether a given country has produced perpetrators of terrorism in one of the periods we are analyzing. 3.2. Methods As we have chosen to measure our dependent variables dichotomously, we rely on a binary logistic regression model in order to asses the impact of political and economic factors on the probability that 1) a terror incident occurs in a country and 2) that a given country gives rise to terrorists. Our basic model then takes the following form: logit (Yi, j ) = α + β1 E + β2 D + β3 Z + εi, j

(2)

where Y is our measure of terrorism (attacks as well as the nationality/origin of perpetrators) in country i at time j; E is a group of economic and institutional variables; D is a measure of democracy, which combines P and C from Equation (1); Z is a group of additional control variables, and εi, j denotes the error term. 3.2.1. The independent variables For both the two applications of our model we use two groups of key explanatory variables. First, we use data from the Economic Freedom of the World Index developed by Fraser Institute and the Economic Freedom Network to measure the kind (and quality) of economic institutions in our sample countries.7 However, consistent with the argument laid out in the theoretical section, we recognize that the impact of state regulations and domestic (economic) institutions on terrorism may be divergent. Rather than employing the aggregate index of economic freedom we therefore use its separate components. The components of the index 6

We have here used the variable concerning the first nationality of the terrorists from the ITERATE database.

7

The most recent edition is Gwartney, Lawson and Gartzke (2005); we have used the data from Gwartney and Lawson (2004). Springer

300

Public Choice (2006) 128:289–315

of economic freedom measure (the absence of) government intervention and the strength of domestic economic institutions in five areas of economic life: (1) the extent of the government spending, taxes and state enterprises; (2) rule of law and the protection of private property rights (i.e. the independence of courts and the ability of courts to enforce property rights); (3) monetary policies (“sound money”); (4) freedom of international trade; (5) the absence of state regulation of labor and credit markets and private business. In our analyses, we have isolated each of these components in order to assess what effect – if any – they have on the probability of terrorist attacks in countries with these properties or that a terrorist has his origin in such a country. Secondly, as a measure of democracy we use Freedom House’s index of democracy (Freedom House, 2004). The Freedom House index consists of two broad measures of democracy: One measuring political rights, i.e. procedural aspects of democracy (such as universal suffrage, electoral competition, etc.), and another measuring civil liberties, i.e. more substantial aspects of democracy (e.g. freedom of speech and organization, etc.).8 Partly because the effects of the various aspects of democracy may have contradictory effects on the likelihood of terror (cf. Davis, 2004; Li, 2005), and partly in order to check the robustness of our results, we use both the combined index of democracy (i.e. political rights plus civil liberties) as well as the individual measures of political rights and civil liberties in separate models. This is also necessary since the two sub-indices are highly correlated with each other.9 Finally, we include a series of control variables in the analyses. These include the squared term of the various democracy indices (in order to allow probabilities to increase/decrease and then decrease/increase); GDP per capita and economic inequality (in order to test for distributional effects, deprivation, modernization, etc.); economic openness/trade; two fractionalization indices (in order to test for the polarization of societies); measures of the religious composition of the population of the countries (in order to test for religious/cultural characteristics), etc. Descriptions of the full set of control variables appear in Appendices A and B.

4. Empirical results Our empirical analyses proceed as follows. We first estimate a series of base line models including only the economic freedom variables and the over-all democracy variable (but not the square terms). Subsequently we estimate a series of expanded regressions including the control variables. For reasons clarified below, we then exclude some of the variables and concentrate on others. These steps are followed for all three periods, i.e. 1996–1999 and 2000–2002 plus the full period (1996–2002).

8 The civil liberties component does, however, also contain some supposed aspects of democracy which in reality are somewhat dubious, e.g., “freedom from gross economic inequality” (Munck & Verkuilen, 2002: 9f). In fact, the entire Freedom House index’s primary value might seem to be that it is the measure of democracy most widely used by researchers (cf. Munck & Verkuilen, 2002: 20f, 27f). 9 In addition and as a further robustness test we have consistently used the co-called Polity II variable from the Polity IV data set as an alternative measure of democracy; the Polity II variable is the combined score of the “Democracy” and “Autocracy” variables from the Policy IV data set (Marshall & Jaggers, 2002). It contains information on aspects such as the competitiveness and openness of executive recruitment and political participation. However, in general the results were fundamentally similar to those resulting from a use of the Freedom House data, and for reasons of exposition the results have not been reported but only been mentioned in the notes when divergent.

Springer

Public Choice (2006) 128:289–315

301

Table 1 Base models: Democracy, economic freedom and probability of transnational terrorism in a country

Size of government Property rights and legal system Sound money Freedom of trade Regulation Democracy (pol.rights + civ.libs.) Constant Pseudo R2 Log likelihood Observations

Model 1 (1996–1999)

Model 2 (2000–2002)

Model 3 (1996–2002)

0.284 (1.50) 0.048 (0.21) 0.131 (1.24) 0.237 (1.08) −0.170 (0.53) −0.025 (2.39)∗ −1.226 (0.84) 0.0973 −69.400 114

0.150 (0.95) 0.172 (0.96) −0.096 (0.75) 0.123 (0.56) 0.364 (1.18) −0.029 (2.97)∗∗ −2.426 (1.66)+ 0.0697 −77.374 121

0.303 (1.44) 0.271 (0.97) 0.108 (0.76) 0.189 (0.67) 0.124 (0.31) −0.048 (3.17)∗∗ −1.784 (0.95) 0.1451 −59.390 114

Absolute value of z statistics in parentheses. + significant at 10%; ∗ significant at 5%; ∗∗ significant at 1%. Method: Binary logistic regression. Dependent variable: Dichotomous terror measure. Economic Freedom variables are measured in 1995 for regressions in 1996–1999. Economic Freedom variables are measured in 2000 for the regressions in 2000–2002. For the period 1996–2002, the average of economic freedoms in 1995 and 2000 is used. Democracy is measured in 1995 for regressions in 1996–1999; and in 2000 for regressions in 2000–2002. For the period 1996–2002, the average of democracy in 1995 and 2000 is used

4.1. Transnational terrorism and freedom in terrorized countries Table 1 shows the results from the base models using only democracy and the economic freedoms as independent variables for the explanation of the occurrence of acts of transnational terrorism. A number of observations are immediately clear from even a cursory glance at the table. First, the economic freedom variables do not display any consistent patterns in relation to terrorism and are always statistically insignificant. Second, only one of the variables has a consistently significant association with the probability that a country will experience terror: The extent of democracy in a country is negatively related to the probability of the same country experiencing transnational terrorism and the coefficients are significant for all three sets of years.10 On the basis of the preliminary results reported in Table 1, we next exclude the economic freedom variables from the regressions, since they have failed to show any significant association with terrorism. Instead we further investigate the association between democracy and terrorism: In Table 2 we add the full set of control variables to the base models of Table 1, including square terms of the various democracy measures in order to test any non-linear relationships between democracy and terrorism. Table 2 displays a number of interesting results vis-`a-vis Table 1. Most importantly a pattern emerges in relation to democracy: Both the over-all democracy index and the sub-index for political rights and the squared terms of these in each instance have coefficients with signs that indicate non-linear relationships between democracy and terrorism, and all the coefficients except one are statistically significant. Specifically, the fact that the over-all democracy and political rights variables are positive while their square terms are negative suggests that the probability of experiencing terrorist incidents is increasing with increasing degrees of 10 When the analyses were run using the Polity variable, the latter was only significant in two of the three periods, while the rest of the results were fundamentally similar to those found when using Freedom House.

Springer

Springer

2.727 (1.91)+ −0.070 (0.73) −0.080 (1.51) −0.027 (3.74)∗∗ 0.000 (0.01) 0.024 (1.27) 0.989 (0.26) −0.036 (1.37) −3.757 (1.87)+ 3.617 (2.06)∗ 0.012 (0.93) 0.018 (1.69)+ 0.003 (0.18) −5.993 (0.89) 0.26 −54.42 109

0.035 (0.89) −0.001 (1.64)+

−0.089 (0.87) −0.082 (1.46) −0.031 (3.86)∗∗ −0.009 (0.30) 0.037 (1.80)+ 2.567 (0.64) −0.047 (1.72)+ −4.349 (2.04)∗ 4.127 (2.19)∗ 0.015 (1.13) 0.020 (1.88)+ 0.002 (0.12) −10.727 (1.50) 0.30 −51.14 109

3.623 (2.40)∗

0.087 (2.28)∗ −0.001 (2.87)∗∗

Model 5 (1996–1999)

−0.023 (0.55) −0.000 (0.13) 1.871 (1.36) −0.081 (0.87) −0.092 (1.75)+ −0.027 (3.69)∗ ∗ 0.010 (0.32) 0.019 (1.02) −0.007 (0.00) −0.029 (1.13) −3.418 (1.76)+ 3.349 (1.99)∗ 0.010 (0.80) 0.014 (1.41) −0.003 (0.23) 1.230 (0.19) 0.24 −55.91 109

Model 6 (1996–1999)

−0.191 (1.53) 0.008 (0.15) −0.022 (2.91)∗ ∗ −0.009 (0.23) 0.011 (0.53) 6.876 (1.61) −0.061 (1.91)+ −3.757 (1.78)+ 4.333 (2.37)∗ 0.020 (1.51) −0.013 (1.02) −0.021 (1.01) −16.034 (2.10)∗ 0.32 −45.66 97

3.035 (2.341)∗

0.113 (1.80)+ −0.001 (2.15)∗

Model 7 (2000–2002)

−0.222 (1.73)+ 0.024 (0.43) −0.023 (3.01)∗ ∗ 0.020 (0.48) 0.016 (0.73) 8.569 (1.92)+ −0.068 (2.03)∗ −4.192 (1.89)+ 5.115 (2.61)∗ ∗ 0.025 (1.76)+ −0.011 (0.88) −0.029 (1.49) −19.272 (2.35)∗ 0.34 −44.32 97

4.264 (2.63)∗ ∗

0.120 (2.32)∗ −0.001 (2.63)∗

Model 8 (2000–2002)

0.054 (0.78) −0.001 (1.22) 3.105 (2.24)∗ −0.162 (1.33) 0.006 (0.11) −0.022 (2.95)∗ ∗ 0.005 (0.13) 0.005 (0.26) 5.441 (1.35) −0.058 (1.89)+ 3.206 (1.62) 3.678 (2.16)∗ 0.017 (1.36) −0.017 (1.41) −0.020 (0.91) −11.519 (1.64) 0.30 −46.85 97

Model 9 (2000–2002)

−0.011 (0.07) −0.082 (1.30) −0.028 (3.13)∗ ∗ 0.004 (0.09) 0.021 (0.94) −2.721 (0.57) −0.047 (1.44) −4.226 (1.94)+ 4.178 (2.17)∗ −0.000 (0.01) 0.007 (0.56) −0.004 (0.24) −13.408 (1.65)+ 0.253 −53.911 113

5.472 (3.06)∗ ∗

0.134 (2.13)∗ −0.002 (2.66)∗ ∗

Model 10 (1996–2002)

−0.035 (0.22) −0.076 (1.09) −0.032 (3.30)∗ ∗ −0.009 (0.22) 0.033 (1.37) −0.375 (0.07) −0.063 (1.65)+ −4.842 (2.04)∗ 4.810 (2.26)∗ 0.002 (0.14) 0.010 (0.85) −0.011 (0.73) −19.141 (2.10)∗ 0.37 −41.83 107

6.580 (3.33)∗ ∗

0.188 (3.11)∗ ∗ −0.002 (3.51)∗ ∗

Model 11 (1996–2002)

0.010 (0.16) −0.000 (0.76) 3.636 (2.38)∗ 0.009 (0.06) −0.094 (1.58) −0.027 (3.23)∗ ∗ 0.030 (0.79) 0.006 (0.31) −4.034 (0.93) −0.040 (1.35) 3.578 (1.78)+ 3.624 (2.04)∗ −0.003 (0.22) −0.001 (0.06) −0.012 (0.71) −3.789 (0.51) 0.26 −49.68 107

Model 12 (1996–2002)

Absolute value of z statistics in parentheses. + significant at 10%; ∗ significant at 5%; ∗∗ significant at 1%. Method: Binary logistic regression. Dependent variable: As in Table 1. Economic freedom and democracy variables enter models as explained in note to Table 1. As for the control variables, they are measured in the following years. For regressions 1996–1999: GDP per cap.: 1995; Growth: average 1996–1999; HPI: 1998; Trade: 1996; Gini; various years (see UNP, 2004); Infant mortality: 1995; Education index: 1994; fractionalization indices: various years (see Alesina et al., 2003). For regressions 2000–2002: GDP per cap.: 2000; Growth: average 2000–2002; HPI: 1998; Trade; 2000; Gini; various years (see UNP, 2004); Infant mortality: 2000; Education index: 1997; fractionalization indices: various years (see Alesina et al., 2003). For regressions 1996–2002: GDP per cap.: average 1995–2000; Growth: average 1996–2002; HPI: 1998; Trade; average 1995–2000; Gini; various years (see UNP, 2004); Infant mortality: average 1995–2000; Education index: average 1994–1997; fractionalization indices: various years (see Alesina et al., 2003); religion variables: mid-1990s

Democracy (pol.rights & civ.libs.) Sq. democracy (pol.rights & civ.libs.) Political rights Sq. Political rights Civil liberties Sq. Civil liberties Log GDP per cap Economic growth Human poverty Trade Inequality Infant mortality Education Latitude Ethnic fractionalization Linguistic fractionalization Muslims Roman-Catholics Protestants Constant Pseudo R2 Log likelihood Observations

Model 4 (1996–1999)

Table 2 Revised models: Democracy and probability of transnational terrorism in a country, with control variables

302 Public Choice (2006) 128:289–315

Public Choice (2006) 128:289–315

303

democracy until a certain point after which the probability of terrorism decreases, i.e., there seems to be a relationship looking like an inverse-shaped U. This is interesting for several reasons. First, it suggests that countries at intermediary levels of democracy are more liable to experience terrorism than purely authoritarian or purely democratic countries. Secondly, it might, but need not, suggest that countries in transition from autocracy to democracy (or the reverse) are more likely to experience terrorism than purely authoritarian or purely democratic countries. The latter two regime forms seem to be the ones that are least likely to experience terrorism, ceteris paribus. This also tells us that there may be nothing unusual about countries (such as Iraq) in transition from one regime type to another experiencing higher level of terrorism. In contrast, having also included squared coefficients for the two components of the Freedom House index we may here observe different a pattern: Civil liberties and transnational terrorism tends to exhibit an association that some times is negative, some times curve-like, but in either case insignificantly so. Other variables in the analysis come out with no or somewhat indeterminate effects. In a number of models higher levels of prosperity (log of GDP per cap.) and lower levels of poverty seem to associate with a higher probability of terrorism, while in contrast more a more tradeoriented economy consistently and strongly associates with less terror – together producing results that are a little difficult to explain. Nonetheless, the latter result is particularly strong and might conceivably be interpreted as an indication that people who trade together do not terrorize each other. It is also interesting that the two indicators of living standards – GDP per capita and the Human Poverty Index (HPI) – seem to have very different associations with the probability of experiencing terrorism. In fact, the HPI is never significant, whereas GDP is significant and positive – i.e. wealthier countries have a higher probability of terrorist acts. So, while there is no support the thesis that poverty causes some countries to experience more terrorism than others, there is confirmation of the claim that wealthier countries are more likely to be hurt. As for the other variables, two fractionalization indices seem statistically significant, but with different effects: Higher ethnic fractionalization, for the most part, decreases the probability of terrorism, while linguistic fractionalization is positively related to the probability of experiencing terrorism. Such results must only be interpreted with the greatest caution, but together they seem consistent with claims that a very culturally diverse society does not itself attract terrorism, whereas societies with sharp divisions along more practical and explicitly nationalist distinctions (such as different languages) may encourage conflict (e.g. in the case of secessionist movements, cf. Muller & Weede, 1990). Another problem may be that ethnic fractionalization and linguistic fractionalization – not surprisingly – are highly correlated (r = 0.70; p < 0.000), which should make us very careful with regard to interpretations of models where they both appear (cf. Alesina, Easterly, Devleeschauwer, Kurlar, & Wacziarg, 2003: 161ff). Geography also contributes to the explanation, in so far as longer distance from Equator tends to reduce the probability of transnational terrorism, but more importantly a number of variables – inequality, economic growth, education, infant mortality and religious composition – show no indications of significantly associating with terror. Together these observations indicate that explanations of terror with reference to relative or absolute deprivation have little to support them.11 11 The results are basically similar, i.e., the coefficients tend to have the same signs and exhibit the same level of significance, when the Polity data are used as when Freedom House variables are used. Among

Springer

304

Public Choice (2006) 128:289–315 Table 3 Revised models: Civil liberties and probability of transnational terrorism in a country

Civil liberties Log GDP per cap. Economic growth Human poverty Trade Inequality Infant mortality Education Latitude Ethnic fractionalization Linguistic fractionalization Muslims Roman-Catholics Protestants Constant Pseudo R2 Log likelihood Observations

Model 13 (1996–1999)

Model 14 (2000–2002)

Model 15 (1996–2002)

−0.028 (1.70)+ 1.816 (1.39) −0.083 (0.91) −0.094 (1.82)+ −0.027 (3.71)∗∗ 0.011 (0.36) 0.019 (1.01) −0.112 (0.03) −0.029 (1.12) −3.400 (1.76)+ 3.335 (1.99)∗ 0.010 (0.79) 0.014 (1.41) −0.004 (0.29) −0.839 (0.15) 0.24 −55.92 109

−0.029 (1.68)+ 2.786 (2.08)∗ −0.135 (1.15) −0.012 (0.25) −0.022 (3.08)∗∗ 0.015 (0.40) 0.001 (0.06) 4.663 (1.22) −0.058 (1.91)+ −2.718 (1.46) 3.601 (2.19)∗ 0.017 (1.34) −0.020 (1.68)+ −0.029 (1.46) −8.178 (1.29) 0.29 −47.62 97

−0.038 (1.83)+ 3.134 (2.32)∗ 0.006 (0.04) −0.100 (1.68)+ −0.027 (3.28)∗∗ 0.036 (1.01) (3.28)∗∗ (0.002) −4.581 (1.06) −0.037 (1.29) −3.421 (1.74)+ 3.495 (2.00)∗ −0.005 (0.36) −0.002 (0.18) −0.015 (1.00) −0.394 (0.07) 0.25 −49.96 107

Absolute value of z statistics in parentheses; + significant at 10%; ∗ significant at 5%; ∗∗ significant at 1%. Method: Binary logistic regression. For explanations on variables and measurement: See notes to Table 1 and 2

Overall, the results from Table 2 lead us to two preliminary conclusions. Firstly, that established negative relationship between democracy (from Table 1) in fact may be less straightforward when other factors are considered. A very undemocratic regime and a very democratic one may both have relatively low probabilities of experiencing terrorist acts, but in between the poles the probability peaks. So, going from low levels of democracy to an intermediary position actually increases the likelihood of terrorism, until some “point of no return” is reached after which the probability of terrorism decreases significantly. Second, a modern, trade-based economy conceivably is a good defense against transnational attacks. Since the values of the civil liberties index do not appear to have a significantly increasing/decreasing effect on the probability of terrorism, we have in Table 3 tested a model in which we exclude the statistically insignificant squared terms of civil liberties (as well as the already excluded but keep the remaining control variables in the model. That is in Table 3 we have revised the models 6, 9 and 12 of Table 2. The results in Table 3 show that better protection of civil liberties tends to significantly associate with a lower probability of transnational terrorist attacks. Societies that protect freedom of speech, freedom of organization etc. – i.e. societies that we tend to see as open and democratic – do not, as some critics would argue and as much anti-terror legislation implicitly assumes, have a higher probability of experiencing terrorism. Quite on the contrary, countries that restrict and suppress civil liberties appear to be more exposed to risks of terrorist attacks.

the few noteworthy differences are that for the sub-period 2000–2002 two of the religious variables come out significantly, when the Polity data are used: the percentage of Muslims in a country is positively associated with the occurrence of transnational terrorist attacks, while the percentage of Protestants is negatively associated with it. Springer

Public Choice (2006) 128:289–315

305

Table 4 Base model: Democracy, economic freedom and probability of terrorist nationality

Size of government Property rights and legal system Sound money Freedom of trade Regulation Democracy (pol.rights & civ.libs.) Constant Pseudo R2 Log likelihood Observations

Model 16 (1996–1999)

Model 17 (2000–2002)

Model 18 (1996–2002)

−0.061 (0.32) −0.218 (0.94) 0.083 (0.75) −0.223 (1.05) 0.458 (1.38) −0.015 (1.62) 0.188 (0.13) 0.07 −67.53 114

0.188 (1.11) −0.139 (0.75) 0.226 (1.50) 0.114 (0.48) 0.033 (0.10) −0.013 (−1.4) −2.921 (1.91)+ 0.07 −70.38 121

0.038 (0.20) 0.086 (0.36) 0.117 (0.90) −0.030 (0.12) 0.173 (0.50) −0.032 (2.90)∗∗ −0.216 (0.14) 0.08 −72.32 114

Absolute value of z statistics in parentheses. + significant at 10%; ∗ significant at 5%; ∗∗ significant at 1%. Method: Binary logistic regression. Dependent variable: Dichotomous measure of nationality of terrorists. For explanations on variables and measurement: See notes to Tables 1–3

The analysis only displays two other consistent relationships: a very significant negative relationship between how trade-oriented a country’s economy is and the probability of terror and a positive relationship between linguistic fractionalization and the probability of terror. 4.2. Transnational terrorism and freedom in terrorist producing countries We now turn to the second half of the analyses, i.e. the investigation of the probability that a given transnational terrorist will have his origin in one country rather than another given the characteristics of the countries. Table 4 contains the results for the logistical regression analysis using the base model, i.e. trying to predict the nationality of the transnational terrorist on the basis of the economic freedom and democracy in his country of origin and with three time periods (1996–1999, 2000–2002 and the full period, 1996–2002). Again, the only independent variable that immediately comes out significant is democracy, i.e., the more democratic a country the smaller the probability that a terrorist will stem from that country (although that result is only statistically significant for the period as a whole). It should be noted that while we have here used the Freedom House democracy measures, we have also run the analysis on the basis of the Polity data; the results were basically identical, although with the latter data some of the economic freedom variables came out with coefficients that were statistically significant, but only at a 10 pct. level, for the sub-period 1996–1999 and with somewhat contradictory signs.12 In Table 5 we again expand the base model to include various control variables and again by considering three time periods. Furthermore, we test models using either Freedom House’s overall democracy index (models 19, 22 and 25) or using the two sub-indices, political rights (models 20, 23 and 26) and civil liberties (models 21, 24 and 27); however, we omit the squared terms for the various democracy variables since these have been found to be insignificant. The results show that adding the control variables increases the explanatory power of the model considerably and consolidates the role of democratic institutions as a significant, negative variable for the prediction of the nationality of a transnational terrorist. In each of 12 Specifically property rights and freedom of trade were, as predicted, negatively related to the dependent variable, while (absence of) regulation was positively related.

Springer

Springer

0.953 (0.68) −0.066 (0.63) −0.048 (0.97) −0.032 (3.16)∗ ∗ −0.025 (0.74) 0.011 (0.57) 1.616 (0.44) −0.022 (0.84) −0.751 (0.46) −0.283 (0.21) 0.008 (0.68) 0.006 (0.68) 0.006 (0.40) 1.241 (0.21) 0.22 −54.88 109

−0.033 (2.27)∗

0.398 (0.30) −0.071 (0.68) −0.047 (0.97) −0.032 (3.17)∗ ∗ −0.022 (0.67) 0.009 (0.46) 1.845 (0.51) −0.025 (0.95) −0.906 (0.56) −0.192 (0.14) 0.011 (0.96) 0.005 (0.53) 0.003 (0.20) 2.520 (0.42) 0.21 −55.95 109

−0.021 (1.82)+

Model 20 (1996–1999)



∗∗

0.137 (0.10) 0.009 (0.08) 0.010 (0.19) −0.033 (3.35)∗ ∗ −0.050 (1.30) −0.003 (0.15) 7.048 (1.79)+ −0.062 (2.05)∗ −1.243 (0.70) 0.280 (0.19) 0.014 (1.16) 0.003 (0.30) 0.008 (0.52) 1.487 (0.23) 0.25 −48.04 97

−0.026 (1.68)+

Model 22 (2000–2002)

significant at 5%;

−0.045 (2.59)∗ ∗ 1.452 (0.99) −0.048 (0.45) −0.048 (0.95) −0.031 (3.11)∗ ∗ −0.028 (0.81) 0.015 (0.77) 1.920 (0.52) −0.021 (0.77) −0.496 (0.31) −0.397 (0.30) 0.007 (0.57) 0.008 (0.81) 0.010 (0.67) −0.549 (0.09) 0.23 −53.92 109

Model 21 (1996–1999)

Absolute value of z statistics in parentheses. + significant at 10%; Tables 1–4

Democracy (pol. rights & civ.libs.) Political rights Civil liberties Log GDP per cap. Economic growth Human poverty Trade Inequality Infant mortality Education Latitude Ethnic fractionalization Linguistic fractionalization Muslims Roman−Catholics Protestants Constant Pseudo R2 Likelihood Observations

Model 19 (1996–1999)

Table 5 Expanded model: Democracy and probability of terrorist nationality

−0.038 (2.06)∗ 0.232 (0.16) −0.004 (0.04) 0.004 (0.09) −0.033 (3.30)∗ ∗ −0.051 (1.29) −0.003 (0.11) 6.871 (1.75)+ −0.060 (1.97)∗ −1.227 (0.69) 0.266 (0.18) 0.015 (1.20) 0.005 (0.43) 0.013 (0.83) 1.818 (0.28) 0.26 −47.26 97

Model 24 (2000–2002)

1.096 (0.89) −0.098 (0.73) −0.091 (1.69)+ −0.029 (3.73)∗ ∗ 0.021 (0.62) −0.023 (1.22) −6.243 (1.48) −0.017 (0.67) −2.822 (1.75)+ 1.165 (0.86) −0.002 (0.13) 0.001 (0.15) −0.001 (0.06) 9.599 (1.64) 0.25 −55.68 107

−0.046 (2.61)∗ ∗

Model 25 (1996–2002)

0.773 (0.65) −0.101 (0.76) −0.088 (1.67)+ −0.029 (3.72)∗ ∗ 0.023 (0.69) −0.024 (1.26) −6.012 (1.44) −0.020 (0.79) −3.004 (1.85)+ 1.229 (0.91) −0.000 (0.04) −0.000 (0.04) −0.005 (0.38) 9.993 (1.72)+ 0.24 −56.23 107

−0.034 (2.50)∗

Model 26 (1996–2002)

−0.050 (2.53)∗ 1.134 (0.91) −0.078 (0.59) −0.087 (1.62) −0.029 (3.66)∗ ∗ 0.018 (0.56) −0.020 (1.06) −5.365 (1.30) −0.017 (0.68) −2.695 (1.69)+ 1.085 (0.81) 0.000 (0.02) 0.002 (0.21) 0.003 (0.18) 8.601 (1.47) 0.24 −55.99 107

Model 27 (1996–2002)

significant at 1%. For explanations on variables and measurement: See notes to

−0.026 (0.02) 0.017 (0.16) 0.015 (0.30) −0.034 (3.37)∗ ∗ −0.050 (1.30) −0.003 (0.14) 7.515 (1.91)+ −0.064 (2.12)∗ −1.262 (0.71) 0.261 (0.18) 0.016 (1.27) 0.002 (0.18) 0.004 (0.29) 1.162 (0.18) 0.24 −48.67 97

−0.016 (1.28)

Model 23 (2000–2002)

306 Public Choice (2006) 128:289–315

Public Choice (2006) 128:289–315

307

the specifications of the model the over-all democracy index is significant and the same goes for the civil liberties index, while political rights only is significant in two of the three time periods. The only other variable that comes out consistently significant (and highly so) is – again – the trade-orientation of a country’s economy, and here the relationship is also negative. However, it might be observed that looking at the full period ethnic fractionalization comes out moderately significant and negative, even when changing the democracy measures.13 In contrast, there is no indication that a country’s economic growth, living standards, distribution of incomes, religious composition, etc., have any consistent, significant effects on the probability that a transnational terrorist will be from that country. In other words: the only thing that we may say with some certainty is that the higher the level of democracy in a country (especially the extent of civil liberties) and the more tradeoriented a country’s economy, the less likely it is that the country will produce terrorists. This may be seen as at least a partial confirmation of the initial hypotheses of the present study.

5. Interpretations and comparisons The results of the preceding statistical analysis have mixed implications for how to interpret the importance of economic freedoms and political freedoms (democracy) with regard to stimulating or discouraging acts of terrorism. Quite fundamentally, economic freedom (or non-freedom) seems in itself to have little or no systematic impact on the occurrence of terrorist acts or in generating terrorists, at least not when the focus is upon transnational terrorism. However, many economists would of course point out that a more trade-oriented economy is far more likely to appear in more free-market oriented societies and therefore that these variables may be somewhat overlapping. In the case of political freedom the results are more mixed. Political freedom seems almost consistently to be of importance for transnational terrorism, i.e., there are indeed associations which are statistically significant. However, the relationships are not always linear and negative and different aspects of political freedom seem to have different results. Specifically, when other factors are considered it seems to be the case that there is a curvelinear, inverse U-shaped relationship between democracy in general (and the extent of political rights in particular) and the probability that a country will experience transnational terrorism. In contrast, the relationship between the extent of democracy in a country and the probability that a transnational terrorist will have his origin in that country is strongly negative. The analysis also indicates that when the components of democracy are broken down, it is political rights which exhibit a curve-linear relationship between democracy and terrorism, whereas civil liberties seems more systematically negatively associated with terrorism. These results may be interpreted as a partial confirmation of the one of the hypotheses of the present paper – that democracy leads to less terror – and one of the possible inferences is that terrorism against the West or political violence within a society does not take place simply because the societies in question are liberal, open and more permissive than the more authoritarian societies.

13 This analysis has also been run with the Polity Democracy-index, which produced quite similar results. The only noteworthy difference in terms of a variable that came out significant using the Polity data which was not so using the Freedom House data was that the percentage of a population that is Muslim was positive for the period 2000–2002.

Springer

308

Public Choice (2006) 128:289–315

But the curve-linear relationship may also be seen as a confirmation of those other theories which see the ability of a regime to suppress rebellions to be important, and that moving from a very authoritarian regime to a half-democratic society may actually unleash forces which previously were kept in control (cf. Tocqueville, [1856] 1955; Davis, 2004). Furthermore, the intermediate levels of political freedom may often be seen in periods of transition, when regimes are weak, and where opposing forces may come out of the closets. Aside from that the perhaps most important and novel finding is that there are consistent, negative associations between how trade-oriented a society is and the probability that it will experience transnational terrorism or generate transnational terrorism. In other words, people who trade a lot seem to be less likely to experience or engage in terrorist acts involving other countries. It also suggests that the positive qualities usually associated with liberal democracies (and emphasized by those who want to export democracy) are not accomplished simply by having elections; other factors, such as civil liberties and a modern, trade-oriented economy may be just as important or perhaps even more so. This in turn suggests that the positive consequences of democracy may come somewhat later than some expect. The present findings must be compared with other, more or less similar results. Specifically, our results with regard to the curve-linear character of the relationship between democracy and terrorism are consistent with a number of relatively recent studies (e.g., Muller & Weede, 1990; Davis, 2004; Abadie, 2004), despite differences in data, time periods and methods. In contrast, our results differ from the empirical analyses of Li, who has recently concluded that “most empirical evidence shows that democracy encourages transnational terrorism” (Li, 2005: 279). Our findings that civil liberties are negatively related to transnational terrorism also goes against the theories and results of Finney and Li (Finney, 1987; Li, 2005). We have not been able to establish the significant negative relationship, which R.J. Rummel has found between various forms of freedom and political violence more broadly conceived (Rummel, 2001: Appendix), although our results with regard to civil liberties are consistent with his (cf. also Rummel, 1985, 1995). Our finding that economic freedom (including the size of the government) is not related in a statistically significant extent to the probability of the occurrence of terrorist acts (at least those of a transnational character) stand in contrast to Finney, who found “the greater the government control over an economy . . ., the less the prevalence of political violence” (Finney, 1987: 206).14

6. Conclusion We have investigated the possible relationships between on the one hand economic freedom and political freedom and on the other hand aspects of transnational terrorism. The hypotheses guiding the analysis have been that more of each of the two types of freedom will tend to be associated with less terrorism – specifically both in terms of the countries in which transnational terrorism takes place and the countries in which border-crossing terrorists originate. Having done so, we have failed to find statistical indication supporting a number of the usually attributed explanations for terrorism, when a number of other characteristics of the particular countries are taken into account: The probability that a country will experience terrorist events is, for example, not statistically significantly related to, e.g., the level of inequality or rapid changes (as in economic growth), etc., and neither is it poor countries that transnational terrorists stem from. 14 Finney’s statistical results are, however, somewhat handicapped by the fact that he omits the then Communist bloc countries from the models including democracy.

Springer

Public Choice (2006) 128:289–315

309

In contrast, the results do seem to lend some support to the hypotheses offered here: There seems, as hypothesized, to be negative relationships between political freedom (in particular civil liberties) and terrorism. However, whereas the relationship between democracy (especially in the form of political rights) in a country and transnational terrorists being generated from that country is clearly negative, there seems to be a non-linear relationship between the extent of political freedom in a country and the probability of the occurrence of transnational terrorist events: Increasing democracy from none to a little may actually increase terrorism, while increasing it to much higher levels eventually will lower terrorism. This observation may be seen as illustrated by cases of countries in the transition from autocracy to democracy (e.g., Iraq and Afghanistan) being marred by, at least in the short run, an increase in domestic political violence. And such, this and the other results also indicate that while democracy may play some role in influencing the occurrence of terrorism, it is not the only one and often far from the most important one. However, there is no observable negative relationship between economic freedom and transnational terrorism and the very first of the hypotheses offered here must therefore be seen as without support.

Appendix A: Data description, variables and definitions

Variable name

Terrorist attack

Origin of perpetrators

Democracy

Variable content

Source

Measurement

Year or period of measurement

Dependent variables ITERATE common file and U.S. Dichotomous: 1996–1999; State Department (Mickolous, 1 = if terrorist 2000–2002; Sandler and Murdock, 2003). event 1996–2002 U.S. State Department occurred. 0 = (ITERATE); (various years) otherwise Same for State Dept. data Nationality of ITERATE common file. Dichotomous: As above perpetrators/first (Mickolous, Sandler and 1 = if person nationality of Murdock, 2003) from country terrorists participates in terrorist event. 0 = otherwise Independent variables Degree of democFreedom House (2004) Recoded to scale 1995 and 2000 racy/autocracy. (1) (www.freedomhouse.org) from 0–100. Freedom House: High values Index (sum of indicate political and civil democracy rights) and separate component values used Occurrence of terrorist incidents. End location of terrorist attack

(Continued on next page) Springer

310

Public Choice (2006) 128:289–315

(Continued)

Variable name

Size of government

Property rights and legal system

Sound money

Freedom of trade

Regulation

GDP per capita

Variable content

Source

(2) Polity IV: Polity Marshall and Jaggers (2002) II variable: SPSS data set and codebook AUTOC variable subtracted from the DEMOC variable Size of government Gwartney, Lawson and Block, spending, taxes (1996), Gwartney and and state Lawson (2003, 2004) Fraser enterprises Institute excel data (Fraser Institute, 2002; and at www.freetheworld.com) Protection of As above property rights by independent courts. Rule of law and ability of courts to enforce property rights Monetary and As above inflationary policies and access to foreign currency Regulation of As above international trade, i.e. tariffs, quotas, capital market controls etc. As above Regulation of labor and capital markets and private business; e.g. extent of restrictions on entry and competition, interest rate control, price controls and government bureaucracy GDP per capita (PPP, Penn World Tables, Mark 6.1. constant 1996 (Heston, Summers and Aten, prices 2002)

Measurement

Year or period of measurement

As above

1995 and 2000

Scale from 1–10; 1995 and 2000 high values indicating freedom

Scale from 1–10; 1995 and 2000 high values indicating freedom

Scale from 1–10; 1995 and 2000 high values indicating freedom Scale from 1–10; 1995 and 2000 high values indicating freedom Scale from 1–10; 1995 and 2000 high values indicating freedom

Scale

1995 and 2000

(Continued on next page)

Springer

Public Choice (2006) 128:289–315

311

(Continued)

Variable name

Variable content

Economic growth

Growth in GDP per capita

Human poverty Trade

Inequality

Infant mortality Education

Latitude

Year or period of measurement

Source

Measurement

Penn World Tables, Mark 6.1. (Heston, Summers and Aten, 2002)

Scale

Human poverty index R.J. Rummel (excel data set from Rummel, 2001), based on UNDP reports Actual trade as share World Development Indicators of GDP [(imports (WDI), World Bank + exports)/GDP] Gini coefficient UNDP (2004)

Scale

1996–1999; 2000–2002; 1996–2002 (for ITERATE). 1996–2003 for State Dept. data 1998

Scale

1995 and 2000

Scale: 0 = total equality of income; 1 = one person receiving all income Scale

1995 and 2000

Infant mortality rate per 1,000 live births UNDP education index

Absolute latitude (distance from equator) Ethnic Ethnic fractionalization fractionalization

World Development Indicators (WDI), World Bank

1995 and 2000

1994 and 1997 Human Development Report, Scale. Index various years (www.undp.org) comprising measures of adult literacy and combined primary, secondary and tertiary enrolment R.J. Rummel (excel data set Scale 2001 from Rummel, 2001) Alesina et al. (2003)

Various years, Scale: 0 = see source homogeneity; 1= heterogeneity. See Alesina et al. (2003) for details (Continued on next page)

Springer

312

Public Choice (2006) 128:289–315

(Continued)

Variable name

Variable content

Year or period of measurement

Source

Measurement

Linguistic Linguistic fractionalfractionalization ization

Alesina et al. (2003)

Religious Religious fractionalfractionalization ization

Alesina et al. (2003)

Muslims

Muslim percentage of population

RomanCatholics

World Christian Database (2005) http://www.worldchristiandatabase.org/ As above

Scale: 0 = Various years, homogeneity; see source 1= heterogeneity. See Alesina et al. (2003) for details Scale: 0 = Various years, homogeneity; see source 1= heterogeneity. See Alesina et al. (2003) for details Scale Mid-1990s

Roman-Catholic percentage of population Protestant percentage As above of population

Protestants

Scale

Mid-1990s

Scale

Mid-1990s

Appendix B. Summary statistics: Independent variables

Democracy, Freedom House 1995 Democracy, Freedom House 2000 Democracy, Freedom House 1995–2000 Democracy, Polity Index 1995 Democracy, Polity Index 2000 Democracy, Polity Index, average 1995–2000 Political rights, Freedom House 1995 Political rights, Freedom House 1995–2000 Political rights, Freedom House 2000 Civil liberties, Freedom House 1995 Civil liberties, Freedom House 1995–2000 Civil liberties, Freedom House 2000 Size of government 1995

N

Minimum

Maximum

Mean

Std. Deviation

188 188 188 157 156 156 188 188 188 188 188 188 121

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.8

100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 9.50

55.6 58.3 56.9 62.1 64.5 63.3 57.1 58.1 59.0 54.1 55.8 57.5 5.6

34.1 33.2 33.1 34.1 33.1 32.6 37.2 36.5 37.3 32.3 30.9 30.4 1.7

(Continued on next page)

Springer

Public Choice (2006) 128:289–315

313

(Continued)

Size of government 2000 Size of government 1995–2000 Property rights 1995–2000 Property rights 1995 Property rights 2000 Sound money 1995 Sound money 2000 Sound money 1995–2000 Freedom of trade 1995 Freedom of trade 1995–2000 Freedom of trade 2000 Regulation 1995 Regulation 2000 Regulation 1995–2000 Log GDP per cap. 1995 Log GDP per cap. 2000 Economic growth, 1996–1999, annual average Economic growth, 2000–2002, annual average Economic growth, 1996–2002, annual average Human Poverty Index 1998 Trade 1996 (% of GDP) Trade 2000 (% of GDP) Trade 1995–2000, average Absolute latitude Education Index 1994 Education Index 1997 Education index, average 1994–1997 Inequality (Gini coefficient) Infant mortality rate 1995 Infant mortality rate 2000 Infant mortality rate, average 1995–2000 Ethnic fractionalization Linguistic fractionalization Percentage Muslim Percentage catholic Percentage protestant

N

Minimum

Maximum

Mean

Std. Deviation

122 120 121 122 122 122 122 121 116 116 122 122 122 121 152 147 148 170 146 187 172 152 169 187 170 173 169 127 193 191 191 185 189 181 201 202

2.3 2.4 2.2 2.2 2.0 0.0 1.5 1.2 1.7 1.7 1.7 2.8 2.7 2.8 2.5 2.5 −5.9 −15.1 −3.5 −10.0 2.6 22.2 2.1 0.0 0.1 0.0 0.1 24.4 3.9 2.9 3.5 0.0 0.0 0 0.01 0.01

9.20 9.35 9.40 9.30 9.60 9.80 9.70 9.70 9.70 9.75 9.80 8.80 8.40 8.55 4.54 4.64 23.53 18.50 18.14 65.00 328.1 341.4 322.3 64.0 0.99 0.99 0.99 70.70 177.00 162.60 165.30 0.93 0.92 99.19 97.76 100.00

5.7 5.7 5.9 5.9 5.8 6.2 7.4 6.9 6.6 6.7 6.9 5.6 5.9 5.7 3.6 3.7 2.2 2.7 2.5 23.5 82.4 91.3 84.4 25.6 0.7 0.7 0.7 40.2 43.56 39.3 41.6 0.4 0.4 23.98 31.22 13.78

1.5 1.5 1.8 1.7 1.9 2.7 2.0 2.2 1.4 1.3 1.2 1.2 1.1 1.1 0.5 0.5 3.2 3.4 2.5 15.4 47.5 50.1 46.7 16.6 0.2 0.2 0.2 10.3 39.42 37.4 38.4 0.3 0.3 35.25 34.68 21.04

References Abadie, A. (2004). Poverty, political freedom, and the roots of terrorism. NBER Working Paper No. 10859. Cambridge: National Bureau of Economic Research. Alesina, A., Easterly, W., Devleeschauwer, A., Kurlat, S., & Wacziarg, R. (2003). Fractionalization. Journal of Economic Growth, 8(2), 155–194. Buchanan, J.M. (1980). Rent-seeking and profit-seeking. In: J.M. Buchanan, R.D. Tollison, & G. Tullock (Eds.), Toward a theory of the rent-seeking society, (pp. 3–15). College Station: Texas A&M University Press. Crain, N., & Crain, W.M. (2006). Terrorized economies. Public Choice, This volume. Davis, G.G. (2004). Repression, rationality and relative deprivation: a theoretical and empirical examination of cross-national variations in political violence. Fairfax: Department of Economics, George Mason University, Mimeo, MS. Springer

314

Public Choice (2006) 128:289–315

DeNardo, J. (1985). Powers in numbers: The political strategy of protest and rebellion. Princeton: Princeton University Press. Drakos, K., & Gofas, A. (2005). Data thinning and the correlation between democracy and transnational terrorism: opening the black-box of under-reporting bias. Patras: University of Patras (Working paper). Finney, L.D. (1987). A rational choice theory of revolution and political violence. Unpublished dissertation. Ph.D. University of Maryland, College Park. Fraser Institute (2002). Economic freedom of the world 1975–2001, Excel data set. Freedom House (2004). Freedom in the world. Excel data set, http://www.freedomhouse.org. Frey, B.S., & Luechinger, S. (2003). Measuring terrorism. IEW Working paper 171. Z¨urich: Institute of Empirical Research in Economics, University of Z¨urich. Frohlich, N., & Oppenheimer, J.A. (1974). The carrot and the stick: optimal program mixes for entrepreneurial political leaders. Public Choice, 19(Fall), 43–61. Gurr, T.R. (1970). Why men rebel. Princeton: Princeton University Press. Gwartney, J.D., & Lawson, R. (2003). The concept and measurement of economic freedom. European Journal of Political Economy, 19, 405–430. Gwartney, J.D., & Lawson, R. (2004). Economic freedom of the World: 2004 annual report. Vancouver: Fraser Institute. Gwartney, J.D., Lawson, R., & Block, W. (1996). Economic freedom of the World: 1975–95. Vancouver: Fraser Institute. Gwartney, J.D., Lawson, R., & Gartzke, E. (2005). Economic freedom of the World: 2005 annual report. Vancouver: Fraser Institute. Heston, A., Summers, R., & Aten, B. (2002). Penn World Table Version 6.1, Center for International Comparisons at the University of Pennsylvania (CICUP), October 2002, excel data set + codebook. Kirk, R. (1983). Political terrorism and the size of government: a positive institutional analysis of violent political activity. Public Choice, 40(1), 41–52. Kotowski, C.M. (1984). Revolution. In: G. Sartori (Ed.), Social science concepts: a systematic analysis, (pp. 403–451). London: Sage. Kurrild-Klitgaard, P. (1997). Rational choice, collective action and the paradox of rebellion. Copenhagen: Institute of Political Science, University of Copenhagen & Copenhagen Political Studies Press. Kurrild-Klitgaard, P. (2004). The paradox of rebellion. In: C.K. Rowley, & F. Schneider (Eds.), The encyclopedia of public choice, vol. 2, (pp. 403–406). Dordrecht: Kluwer. Li, Q. (2005). Does democracy promote or reduce transnational terrorist incidents? Journal of Conflict Resolution, 49(2), 278–297. Lichbach, M.I. (1987). Deterrence or escalation? The puzzle of aggregate studies of repression and dissent. Journal of Conflict Resolution, 31(2), 266–297. Lichbach, M.I. (1995). The rebel’s dilemma. Ann Arbor: University of Michigan Press. Lichbach, M.I., & Gurr, T.R. (1981). The conflict process: A formal model. Journal of Conflict Resolution, 25, 3–29. Marshall, M.G., & Jaggers, K. (2002). Polity IV project: Political regime characteristics and transitions, 1800–2002. Data set and codebook, http://www.cidcm.umd.edu/inscr/polity/. McCormick, G.H. (2003). Terrorist decision making. Annual Review of Political Science, 6, 473–507. Mickolous, E.F., Sandler, T., & Murdock, J.M. (2003). International Terrorism: Attributes of Terrorist Events, 1968–2003 (ITERATE). Dunn Loring: Vinyard Software. Muller, E.N., & Weede, E. (1990). Cross-national variation in political violence: A rational action approach. Journal of Conflict Resolution, 34(4), 624–651. Munck, G.L., & Verkuilen, J. (2002). Conceputalizing and measuring democracy: Evaluating alternative ideas. Comparative Political Studies, 35(1), 5–34. Olson, M. (1971 [1965]). The logic of collective action: Public goods and the theory of groups, 2nd edn. Cambridge: Harvard University Press. Ostrom, E. (1986). An agenda for the study of institutions. Public Choice, 48, 3–25. Rathbone, A., & Rowley, C.K. (2002). Terrorism. Public Choice, 111, 1–10. Rosendorff, B.P., & Sandler, T. (2005). The political economy of transnational terrorism. Journal of Conflict Resolution, 49(2), 171–182. Rummel, R.J. (1984). Libertarianism, violence within states, and the polarity principle. Comparative Politics, 16, 443–462. Rummel, R.J. (1985). Libertarian propositions on violence within and between nations: A test against published research results. Journal of Conflict Resolution, 29(3), 419–455. Rummel, R.J. (1995). Statistics of democide. New Brunswick: Transaction. Rummel, R.J. (2001). Saving lives, enriching life: Freedom as a right and a moral good. Manuscript, http://www.mega.nu:8080/ampp/rummel/. Springer

Public Choice (2006) 128:289–315

315

Shughart, W.F. (2002). September 11, 2001. Public Choice, 111, 11–18. Tocqueville, A.D. (1955 [1856]). The old regime and the French revolution. New York: Doubleday. Tullock, G. (1971). The paradox of revolution. Public Choice, 11, 89–100. Tullock, G. (1974). The social dilemma: The economics of war and revolution. Blacksburg: Center for Study of Public Choice. Tullock, G. (2004 [1967]). The welfare costs of tariffs, monopolies, and theft. In: C.K. Rowley (Ed.), Virginia political economy, The selected works of Gordon Tullock, Vol. 1, pp. 169–179. Indianapolis: Liberty Fund. Tullock, G. (2005). The rent-seeking society. In: C. K. Rowley (Ed.), The selected works of Gordon Tullock, Vol. 5. Indianapolis: Liberty Fund. UNDP (1996–2004). Human Development Report. www.undp.org. U.S. State Department (1996–2003). Patterns of Global Terrorism (1996–2003). http://www.state.gov/s/ct /rls/pgtrpt/. World Bank. World Development Indicators. CD-ROM and Online, Washington. World Christian Database (2005). http://www.worldchristiandatabase.org/.

Springer