ANNUAL BANK CONFERENCE ON ... - World Bank Group

5 downloads 0 Views 301KB Size Report
Second, studies of civil war find that highly diverse societies face a lower risk of civil war ..... These data have their origin in the U.S. Department of State's World Military Expenditures ..... the regional classification follows that of World Bank (2004). .... Making Peace Pay: A Bibliography on Disarmament and Conversion.
Annual Bank Conference on Development Economics Amsterdam, The Netherlands May 23-24, 2005

Disarming Fears of Diversity: Ethnic Heterogeneity And State Militarization, 1988 - 2002 Indra de Soysa Norwegian University of Science and Technology (NTNU)

The World Bank

Disarming Fears of Diversity: Ethnic Heterogeneity and State Militarization, 1988–2002 ∗

Indra de Soysa Dept. of Sociology and Political Science Norwegian University of Science and Technology (NTNU) & Eric Neumayer London School of Economics and Political Science Department of Geography and Environment

March 2005



An earlier version was presented at 46th Annual International Studies Association Conference in

Hawaii. We would like to thank Jim Fearon and Erich Weede for many helpful comments. Eric Neumayer acknowledges financial assistance from the Leverhulme Trust.

3

Disarming Fears of Diversity: Ethnic Heterogeneity and State Militarization, 1988–2002 This study addresses state militarization under conditions of ethnic, linguistic and religious diversity. We begin on the basis of two empirical findings. First, recent scholarship in economics finds that a high level of diversity leads to lower provision of public goods. Second, studies of civil war find that highly diverse societies face a lower risk of civil war despite the ‘primordialist’ claims about ancient hatreds, fear, and insecurity in such societies. We investigate whether there is a link between these two separate lines of research. Could it be that diversity prompts governments to militarize heavily in order to prevent armed conflict, which would then crowd out other spending on public goods? Using military expenditures, military personnel and arms imports as indicators of militarization, we find that higher levels of ethnic (and possibly linguistic) diversity predict lower levels of militarization, whereas religious diversity does not matter. If high diversity renders civil war more unlikely, then it does not happen via a ‘garrison state’ effect. If diverse societies spend less on public goods, then this is not because they are crowded by spending on militarization.

4 There are two important strands of theoretical and empirical scholarship on the effects of ethnic, linguistic and religious diversity on state behaviour. First, scholars interested in governance and public spending find that heterogeneity leads to lower provision of public goods, such as education, health, and infrastructure. Since diversity poses problems for arriving at a consensus for co-operative solutions (a question of governance under diverse preferences), the greater the diversity the worse the policy outcomes (Alesina, Baqir, and Easterly 1999; Alesina et al. 2003; Easterly 2001). Secondly, the nature of many of the armed conflicts following the end of the Cold War refocused attention on the consequences of ethnicity, religion, and other forms of cultural heterogeneity1 on social outcomes (Alesina et al. 2003; Collier 2001a; Ellingsen 2000; Fearon and Laitin 2003; Fox 2004; Gurr 1993; Harrison and Huntington 2000; Horowitz 1998; Ramet 2004; Reynal-Querol 2002; Sambanis 2001; Varshney 2001; Wimmer et al. 2004). 2 The focus on ethnicity and religion surely intensified following the September 11th terrorist attacks in the US. Yet, the tradition of explaining Third World violence in years after World War II as ethno-nationalist rebellion under conditions of weak states has deep roots (Drake 1957; Gurr 1970; Huntington 1968). Ethnic and, if less so, religious conflict is supposed to be ‘endemic’ and ‘everywhere on the rise’. 3 However, contrary to the conventional wisdom, ethnic and religious fractionalization

1

In the following we use the terms heterogeneity, diversity and fractionalization interchangeably.

2

It is reported that one English-language scholarly journal database records 249 articles published since

1990 containing ‘ethnic conflict’ in the title as opposed to just 23 with ‘class conflict’ (Gilley 2004: 1155). Surprisingly, ethnicity has never entered models explaining militarization. 3

See Kaplan (1994) for a recent explication of the primordialist argument that suggests ethnic and, more

broadly, cultural conflict to be endemic. Others argue that the spread of free markets and democracy increase ethnic conflict (Chua 2003). Huntington’s (1993, 1997) hypothesis of a Clash of Civilizations provides a related argument. Yet, many find little empirical support for civilizational bases for armed

5 does not seem to lead to a higher risk of war onset if one looks at all types of civil war together (Fearon and Laitin 2003; Mueller 2000). If anything, high diversity makes countries safer, or in other words, ethnic dominance (Collier 2001a; Collier and Hoeffler 2004a) or ethnic polarization (Montalvo and Reynal-Querol 2005) is what matters, not ethnic fractionalization. suggest dropping the concept from academia altogether (Gilley 2004). Nonetheless, theories built around the concepts of ethnicity, religion and security dilemmas are particularly prominent in the literature (Kaufmann 1996; Petersen 2001; Posen 1993; Snyder and Jervis 1999; Walter and Snyder 1999). Ethnic and religious conflict occurs because groups are unable to coordinate mutual security fears, particularly over questions of national integration, leading to misperceptions, emotion, rage, violence etc. 4 Could it be that there is a link between the two literatures? This is what this paper’s analysis tries to explore. Ethnic fractionalization might lead to security concerns to which governments respond with higher militarization. Since fractionalization might pose a readily observable security threat, perhaps governments over-compensate the security risks, neglecting other public goods? Could it be that governments try to ensure peace and security under conditions of high diversity through militarization at the cost of other public goods? In other words, does militarization crowd out the provision of non-security public goods in a special kind of guns-versus-butter trade-off? Or does diversity reduce militarization, which might follow from the fact that security also represents a form of public good, the governmental provision of which might be subject to the same diminishing effects of heterogeneity? Thus, the question we address directly is what effect does ethnic, linguistic

violence (Henderson 1997; Russett, Oneal, and Cox 2000), and for claims that the incidence of cultural conflicts is increasing (Marshall and Gurr 2003). 4

For evaluation of theories explaining the Yugoslav meltdown, see Ramet (2004). Political scientists

have been preoccupied with problems of pluralism and problems of political integration.

6 and religious diversity and polarization have on a state’s extent of militarization? We test this proposition employing several measures of diversity and polarization on three indicators of state and social militarization reflected in military expenditures, the share of military personnel in the population, and arms imports. Our results are easily summarized. Using the latest military expenditure data from the World Bank (2004), we find that ethnic heterogeneity predicts lower rather than higher levels of military spending to GDP between 1988 and 2002, controlling for several salient factors, such as country size, income, regime type, security risks, armed conflict etc. If states fear ethnic heterogeneity, it does not show in terms of how they prepare to deal with it. The result is robust to sample size and several different specifications and testing procedures. Ethnic heterogeneity is also negatively related to the share of military personnel in the total labor force. Since most poor countries are likely to follow more labor-intensive militarization strategies, the negative relationship between heterogeneity and the share of military personnel in the labor force is also instructive. Religious heterogeneity had no statistically significant effect in any of the tests, which confirms existing studies that fail to find an effect of religious heterogeneity on either growth or institutional quality. In one set of estimations, it seems to be linguistic rather than ethnic heterogeneity that diminishes militarization, but Alesina et al.’s (2003) linguistic fractionalization measure is highly correlated with ethnic fractionalization as measured by others. We fail to find evidence that it is ethnic and religious polarization rather than fractionalization that matters, as some have argued (e.g., Montalvo and Reynal-Querol 2005). The results taken together do not suggest that governments run scared because of ethnic and other diversity – quite the opposite. The results do not support those who contend that ethnic diversity is predisposed to spur militarization due to security fears. Similar to the problems diverse societies encounter for the provision of public goods, it seems that they may have difficulties collecting the necessary taxes, forging necessary political support, or

7 reaching the necessary social consensus required for militarization. This does not mean we would advocate militarization as a solution, but rather that governments do not seem to act the way we think would be a rational response to real and/or perceived threats emanating from ethnic diversity. Our results also throw into doubt the notion that minority ethnic groups need to fear high state militarization when they rebel for autonomy. Neither do the results support a conjecture that heterogeneous societies remain peaceful because states militarize to prevent violent conflict. Realist theories in particular argue that ethnic conflict in Eastern and Central Europe was kept in check by Soviet military might, only to erupt with the withdrawal of Soviet power (Huntington 1993; Mearsheimer 1990). If in fact diversity is a source of potential violent conflict, it does not seem likely that peace prevails purely because of a ‘garrison state’ effect. We find exactly the opposite of this expectation regarding state behavior under conditions of ethnic and other diversity. Nor does it seem that ethnicity prevents higher provision of education and other public goods because military budgets crowd out public goods. What are the implications of our results on the merits of diverse as opposed to homogeneous societies? This depends on how one regards militarization. Looked at as a measure for the provision of the public good security, militarization is likely to be underprovided and the fact that fractionalization leads to even lower provision does not bode well for diverse societies. However, if one believes with Kant (1795) and other liberals that military expenditures are likely to be excessive and indeed often dangerous as they spur rather than deter conflict and lower rather than increase security (Collier and Hoeffler 2004a; 2004b), then this speaks in favor of diverse societies. In our own view, which concurs with that of Collier (2001) and others, greater diversity is a source of good, which means that encouraging separatism for ending ethnic enmities might be counter-productive in the long run.

8 Ethnic Divisions, Preferences, and Public Goods Ethnic and other forms of diversity are not interesting only because of their supposed links to violent conflict. Recent research finds that it also affects economic development and other public policy outcomes because diversity makes consensus difficult. Ethnic heterogeneity (and polarization) is seen as the underlying cause of the failure of collective action, particularly as it generates incentives for rent-seeking (Alesina 1994; Alesina and Drazen 1991; Alesina and Rodrik 1994; Garcia-Montalvo and Reynal-Querol 2005). Political economy models suggest that heterogeneity is “prone to competitive rent-seeking by the different groups that have difficulty agreeing on public goods like infrastructure, education, and good policies” (Easterly and Levine 1997: 2). This phenomenon has been demonstrated at various levels of aggregation – see, for example, Alesina, Baqir and Easterly’s (1999) study of the negative impact of ethnic fractionalization on public good spending in U.S. cities and Easterly and Levine’s (1997) cross-national study explaining a good part of Africa’s growth tragedy with the negative effects of its high degree of ethno-linguistic fractionalization on political stability, the provision of public goods and growth-promoting policies. 5 Apparently, the diverse preferences of heterogeneous groups make it more difficult to forge public policies that benefit the collectivity. In the US city example, it is noted that white Americans prefer low taxes in cities with heterogeneous populations because of the belief that blacks benefit predominantly from spending on public goods. Africa’s economic woes are directly related to the bad public policy decisions made as a result of ethnic heterogeneity, where political conflict impedes public good provision such as education and health (Easterly and Levine 1997). Other cross-national studies show that ethnic polarization 5

Daniel Posner corrects Easterly and Levine’s (1997) measure for ethnic groups that are politically

relevant and comes to the same conclusion (Posner 2004). Alesina et al. (2003) have revisited the issue with indices measuring ethnic, linguistic and religious fractionalization separately and find that it is ethnic and linguistic fractionalization that matters, but not religious fractionalization.

9 lowers investment, whereas religious polarization increases government consumption relative to GDP (Garcia-Montalvo and Reynal-Querol 2005). However, Alesina et al (2003) find that it is diversity that matters on the question of poor economic policy and public goods provision, not polarization. Possibly, the negative effects of fractionalization are mitigated in democracies (Collier 2001) or rather, as Easterly (2001) argues, where institutional quality is high, which is only weakly correlated with democracy. However, institutional quality is itself negatively affected by ethnic fractionalization (Alesina et al. 2003; Keefer and Knack 2002). 6 Recent studies of civil war show that ethnic diversity’s role in conflict is not straightforward. Ethnicity is important of course for organization and mobilization of support, but conflict occurs when the opportunity for using violence is maximized. Some studies have shown a curvilinear association between ethnic fractionalization and conflict. Highly homogeneous and highly heterogeneous societies are both able to maintain peace. In highly homogeneous societies there is little ethnic strife, whereas a high degree of fractionalization prevents mobilization on issues. Collier (2001) claims that highly fractionalized societies will pose difficulties for large enough minimum winning coalitions to form so that groups will not be able to challenge a states’ monopoly on force effectively. The trouble is in between with moderately fractionalized societies facing the greatest danger (Collier and Hoeffler 2004a; de Soysa 2002). Others call this polarization, where two equally sized groups are the most dangerous, or in other words, where moderate fractionalization prevails, since measures of polarization are at a maximum when society is made up of two groups containing 50% of the population each (Alesina et al. 2003; Garcia-Montalvo and Reynal-Querol 2002). Moreover, if the largest minority is large enough, it is a more attractive target for expropriation by a majority, thus leading to polarized conflict and violence. Yet, some could possibly claim that

6

The bivariate correlation between the International Country Risk Guide’s (ICRG) institutional quality

measure and Fearon’s (2003) measure of ethnic fractionalization is –0.32 in our sample.

10 the curvilinear relationship is a result of the ways in which states avoid militarized conflict. For example, if states anticipate a high probability of violence or ethnic challenges under conditions of heterogeneity, they could more effectively deter it through militarization, whereas less fractionalization may lead states to neglect security. It is this aspect of the debate we test on militarization, only discussing the question of violent conflict as a backdrop for why militarization should matter. Ethnic fractionalization and polarization are also supposed to lead to violence because of the ways in which state power is distributed. If a majority hogs state power, then it has an incentive to militarize in order to prevent a challenge from other groups. This would lead to cycles of repression, causing security dilemmas (Kaufmann 1996; Posen 1993; Snyder and Jervis 1999). In fact, the idea of an ethnic security dilemma borrows heavily from the international relations literature where states operating in an anarchical international system are driven to seek security, but create insecurity among others in so doing. 7 The concept of an ethnic security dilemma is applied in the situation where states have collapsed and ethnic groups apparently go from cooperation to violence because of mutual fears about the intentions of others. However, the reasons for state collapse and the events that drive the security dilemma are artificially separated in this literature (Roe 2000). In any case, it is not just the rhetoric of nationalist leadership that ethnic groups and others fear, but also their actions, which are almost always explicitly demonstrated by military means. The military is also an important source of patronage and power. Powerful groups can use ethnic nepotism to gain from controlling military purse strings. Some report that corruption increases military

7

In fairness to theories of the security dilemma and ethnic conflict, most suggest that a period of

anarchy, i.e. state collapse, precedes the security dilemma. However, these situations are extremely rare. Also, could it not be that those factors that led to state collapse in the first place also account for why ethnic cooperation breaks down (Roe 2000)?

11 spending (Gupta, de Melo, and Sharan 2001) and others that ethnicity increases corruption (Alesina et al. 2003; La Porta et al. 1998), although no one reports the effects of diversity on military spending itself. Thus, ethnic conflict could very well be one aspect of a cocktail that includes ethnicity, corruption, and militarization. Ethnic conflict also relates importantly to the ways in which colonial powers used the tactic of ‘divide and rule’ to control colonies, such as in Rwanda, where a minority was given control of military matters and the power to tax. For our purposes, however, we do not think it is important as to whether a minority holds state power or a majority, but that the logic of the security dilemma should apply equally in either case. We simply observe patterns of military spending under conditions of ethnic heterogeneity by governments. Our study is motivated by at least two interrelated concerns. The first is theoretical from the perspective of conflict studies. If ethnic diversity is inherently dangerous, then do states prepare to meet it via militarization? If ethnic heterogeneity does not seem to matter in terms of the outbreak of civil war is this because states suppress conflict effectively by increasing military capacity? In other words, does latent conflict prompt state militarization, particularly since as some claim ‘ethnic conflict is a tragic constant of human history’ (Easterly 2001: 687)? Secondly, the question of ethnic heterogeneity and militarization is prompted by what seems to be a strong empirical association relating ethnic divisions to lower levels of public good provision in general. Thus, do states facing potential ethnic conflict spend on militarization at the expense of other public goods? In fact, several scholars treat military spending as a public good both regionally and within countries because if in fact it buys security, then others benefit from having to spend less given the regional nature of the consequences (Collier and Hoeffler 2002; Olson 1982). In other words, since diversity is readily observable (the tragic constant) and states could prepare proactively to prevent the supposed dangers from diversity through militarization, does military spending crowd out other public goods? The question we focus on here is therefore: if diversity is bad for the

12 provision of public goods such as infrastructure and education, is this because heterogeneity requires a greater share of a society’s income to be devoted to the public good of security through militarization?

Methods & Data We employ a pooled time-series, cross-section (TSCS) data set. 8 Our main dependent variable consists of military expenditures over GDP, which the literature usually refers to as the defense burden. 9 We keep this variable in its level form, but our main results are hardly affected if it is logged instead (Collier and Hoeffler (2004b) come to the same conclusion in their study). The data are taken from the World Bank’s World Development Indicators (WDI) CD-Rom (World Bank 2004), which is also the source for the other variables unless noted otherwise. They are available annually from 1988 up to 2002, a total of 15 years, but the first year is lost due to the use of a lagged dependent variable. Combining various sources one could in principle construct a panel that reaches further back in time. However, given measurement and international and inter-temporal comparability problems with military expenditure data particularly during the period of the Cold War (Brzoska 1995), we prefer to 8

The entire data set and do-files will be made available upon publication.

9

World Bank (2004) provides the following definition: “Military expenditures are based on the NATO

definition, which includes all current and capital expenditures on the armed forces, including peacekeeping forces; defense ministries and other government agencies engaged in defense projects; paramilitary forces, if these are judged to be trained and equipped for military operations; and military space activities. Such expenditures include military and civil personnel, including retirement pensions of military personnel and social services for personnel; operation and maintenance; procurement; military research and development; and military aid (in the military expenditures of the donor country). Excluded are civil defense and current expenditures for previous military activities, such as for veterans' benefits, demobilization, conversion, and destruction of weapons.”

13 use one single data source that largely covers only the post-Cold War period. This is of course also the period in which ethnic conflict is supposed to be particularly prevalent. The recent data are also much more reliable in terms of greater levels of transparency within much of the developing world (Omitoogun 2003). The World Bank data originally come from the Stockholm Peace Research Institute’s (SIPRI) series of Yearbook: Armaments, Disarmament and International Security and are almost identical to data supplied to us directly by SIPRI (r = .98). In addition, we use two other variables capturing different aspects of militarization to check the robustness of our results. The second dependent variable is military personnel as a share of the total labor force, which has slightly lower data availability compared to the military expenditure data. The advantage of using the share of military personnel as well as military spending is that poor countries may simply use labor-intensive (rather than capitalintensive) forms of militarization. Lastly, we also use a variable measuring arms imports relative to total imports. This variable is only available up to 1999 and in principle arms import expenditures should be included in total military expenditures. However, for some countries arms imports are not accounted for in military expenditures (Brzoska 1995) and a high arms to total imports ratio provides yet another feature of a highly militarized society. These data have their origin in the U.S. Department of State’s World Military Expenditures and Arms Transfers reports and are taken from World Bank (2003). 10 Our primary independent variable is ethnic heterogeneity, which has received a great deal of attention in the recent empirical literature in several fields, particularly in political 10

The underlying definition is as follows: “Arms imports are imports of military equipment usually

referred to as ‘conventional’, including weapons of war, parts thereof, ammunition, support equipment, and other commodities designed for military use.” There is one data point of more than 100 per cent (Ethiopia 1989), which can happen if there are inconsistencies in the reporting and measurement of arms as well as total imports. Dropping this observation from the sample had little impact on the results.

14 science, sociology, and economics (Alesina et al. 2003; Fearon 2003). As a result, the measures have improved tremendously over the one previously developed by Soviet social scientists in the 1960s. Defined as the probability that two randomly selected individuals from the same country belong to different ethnic or linguistic groups, heterogeneity is n

computed as ELF ≡ 1 − ∑ pi2 , where pi is the population share of ethnic or linguistic group i i =1

and n is the number of existing groups. This measure of ethnolinguistic fractionalization (ELF) mixed ethnic and linguistic characteristics. Recently, however, Alesina et al (2003), Montalvo and Reynal-Querol (2005), Fearon and Laitin (2003) and Fearon (2003) independently constructed new measures of ethnic and religious fractionalization (as well as linguistic fractionalization in the case of Alesina et al. 2003). They use the same formula, but improve upon the older data significantly. These data series are based on more current, updated sources, and do not conflate ethnic and linguistic characteristics in a single measure as blatantly as the old ELF measure. They also rely on survey-based studies that have examined several African countries where distinction of groups is not always straightforward. While there are several definitional and operational differences in their methods, Fearon (2003: 196) claims that his measure is ‘broadly similar’ to Alesina et al’s (2003) measure. Both these measures are highly similar to Montalvo and Raynal-Querol’s (2005). In addition to these three sources, two other scholars have also provided indices of fractionalization that are perhaps less well known and that do not distinguish ethnic from religious fractionalization. Phillip Roeder bases his ethnolinguistic fractionalization index mainly on the original Soviet sources from the 1960s together with other Soviet ethnographic studies from the 1980s (Roeder 2001). Yeoh Kok Kheng provides yet another measure of ethnic fractionalization, based on racial (phenotypical), linguistic and religious differences, using a large variety of sources (Kheng 2001).

15 We primarily rely on Fearon’s (2003) and Fearon and Laitin’s (2003) widely used measures of ethnic and religious fractionalization. We also use the alternative measures to test the robustness of our results, a strategy advocated by both Fearon (2003) and Alesina et al. (2003). Additionally, Montalvo and Reynal-Querol (2005) argue that it is polarization rather than fractionalization that matters. Their formula for polarization is much more complex than the formula for fractionalization. It measures “the normalized distance of a particular distribution of ethnic and religious groups from a bimodal distribution” (see Montalvo and Reynal-Querol 2005: 301 for details). Polarization approaches unity when the population is made up of two equally sized groups and then declines as the number of groups increases further, whereas fractionalization increases monotonically with the number of groups. Empirically, across countries ethnic polarization is related to ethnic fractionalization in a non-linear way: Ethnic polarization first rises with increasing fractionalization, but then falls at an intermediate level of fractionalization. Religious polarization is somewhat different; it first increases as fractionalization increases and then at higher levels of fractionalization there is no relationship to polarization. See Montalvo and Reynal-Querol (2005) for a detailed discussion. Thus, we also test ethnic and religious polarization in our models of militarization. Table 1 provides a correlation matrix for the various measures of fractionalization and polarization used. There is clearly often strong correlation among the various measures, but they are far from identical. We conclude that it is important to test the various available measures if we want to ensure robustness of our results. Turning to control variables, there is an enormous theoretical and empirical literature that has accumulated on the causes and consequences of military spending (Gleditsch et al. 2000; Hartley and Hooper 1990). Most of these studies have focussed on arms races between the superpowers, or are case studies of single countries over time. We rely primarily on two recent empirical studies addressing the determinants of military spending (Collier and Hoeffler 2002; Goldsmith 2003), none of which addressed ethnic and other diversity,

16 however. Collier and Hoeffler (2004b) use five-year averages from 1960-1999. Their objective was to see what drove military budgets and how military expenditure affected conflict. They did not include a lagged dependent variable, which is one of the main ways their study differs from ours. Goldsmith (2003) in his recent empirical study on the determinants of military spending that tested at least 18 theoretically informed variables on defence spending between 1886 and 1989 found support largely for domestic factors explaining military spending, such as income per capita and democracy. This corroborates the findings in the Collier and Hoeffler (2004b) study using a different operationalization. International variables, such as level of conflict in the neighbourhood, extended rivalries, alliances, and levels of military spending in the neighbourhood had only weak support compared with the domestic factors. However, Goldsmith (2003), like Collier and Hoeffler (2004b), found positive effects of involvement in civil and or international war on higher military spending as expected. Importantly, like Goldsmith (2003) we control for the previous year’s value of the dependent variable. To start with, this is foremost for theoretical reasons. It is very likely that military budget decisions are subject to bureaucratic inertia (Goldsmith 2003; Gupta, de Melo, and Sharan 2001) due to the presence of entrenched interests or what the late US President Eisenhower called the influence of the military-industrial complex. As Goldsmith (2003) argues, not controlling for the inertia in spending levels in many previous studies should lead us to treat with caution several claims about the causal relationship of some variables highlighted in the literature. But there are also practical reasons for including the lagged dependent variable. First, its inclusion controls for temporal dependence and serial

17 correlation (Beck and Katz 1995b). 11 Second, the lagged dependent variable is highly correlated with any potentially omitted variable such as the perceived threat to the security of the country, which is difficult to model explicitly and comprehensively. We also control for level of per capita income as well as its growth rate, which are commonly used variables (Davoodi et al. 2001; Goldsmith 2003; Gupta, de Melo, and Sharan 2001). We use gross national income per capita in purchasing power parity (PPP). This variable is logged to reduce skewness. Most find that income is positively related to higher expenditures, arguing that wealth allows governments the greater luxury of stronger defense (Collier and Hoeffler 2002). In economic terms, military spending is likely to be a normal good, that is a good with a positive income elasticity (Sandler and Hartley 1995). High economic growth rates might make it easier for governments to impose a greater defense burden on society. We use total population (logged) to control for country size because this influences both ethnic heterogeneity and militarization. Collier and Hoeffler (2004b) report a negative effect of country size (population) on military budgets, arguing that large countries deter external threats. We control for regime type using the POLITY IV dataset’s polity2 indicator, which uses a weighting scheme to treat periods of transition 12 . We expect autocracies to have higher military spending than democracies (Collier and Hoeffler 2002; Goldsmith 2003). Many have argued that autocracies are dependent on military force to sustain their rule against attempts to overthrow their government, whereas democracies command a greater degree of legitimacy and are thus less in need of a strong military (Kimenyi and Mbaku 1995; Maizels and Nissanke 1986). In addition, heavy militarization 11

A lagged dependent variable can potentially mask any causal effects explaining by the other

explanatory variables. Thus, our tests are extremely conservative and any significant effects are bound to be robust, and the reported substantive effects possibly smaller than in reality (Achen 2000). 12

The data and explanations can be obtained from http://www.cidcm.umd.edu/inscr/polity/ (accessed

December 2004).

18 might represent a threat to democratic rule as it makes it easier for the military to influence politics to the point of removing the democratically elected leaders altogether. We additionally control for overall government spending, since high government consumption generally will have the same causes as high military consumption. Next, we control for internal and external security threats as militarization variables will be driven by such factors (Collier and Hoeffler 2002). We enter a term for civil war, which is a dummy variable for years in which a country experiences armed conflict with over 25 battle-related deaths (Gleditsch et al. 2002). Following Goldsmith (2003), the international war variable is a dummy for years in which a country engages in conflict between states with at least 1000 deaths. These data are taken from Gleditsch et al. (2002). We also compute a count of civil and international peace years, or the number of years since the last civil and international war since 1946 in order to gauge the proximity of previous conflict (Collier and Hoeffler 2002). It is well established that, for civil wars at least, there is a high risk of revival even after a civil war has formally ended, which suggests that militarization after the end of civil war is likely to diminish only slowly over time (Collier and Hoeffler 2004b). We run tests with and without the civil war variables (incidence and peace years) because these conflicts could be purely related to ethnic heterogeneity, in which case they explain all the effects of government response to militarization. They could also be endogenous to militarization. Similar to Collier and Hoeffler (2004b) we take a weighted average level of militarization of countries that are “contiguous”. The weight is GDP and contiguity is defined as either land contiguity or water contiguity up to 400 miles of water. Data are from the Correlates of War (COW) project and were taken from Bennett and Stam (Bennet and Stam 2003). In a context of rivalry, the militarization level of contiguous countries can capture local arms race phenomena. In a context of non-rivalry, it can capture emulation, imitation and coordination effects. Contrary to Collier and Hoeffler (2004b), we do not include a

19 measure of predicted civil war. Such a variable creates all kinds of statistical problems. Instead, we control for the risk of civil war directly by our range of explanatory variables together with the lagged dependent variable, which will capture the risk of civil war under the assumption that the factors triggering such war are time-persistent. Finally, we include yearspecific dummies to capture any trends over time and year-specific international tension that influences defense spending globally, such as the end of the Cold War, the Persian Gulf War, or NATO action in the Balkans. Table 2 provides summary descriptive variable information. The estimation of TSCS data presents some special problems, particularly because of complex correlation patterns between and across panels (Beck and Katz 1995a; Beck and Katz 1995b). The problems relate to the way in which standard errors are computed and problems of serial correlation and heteroskedasticity. Since our data is unbalanced to an extent that no time periods are common to all countries in the sample, the standard version of the Panel Corrected Standard Errors (PCSE) method of Beck and Katz, which assumes that panels are not only heteroskedastic, but also contemporaneously correlated across panels, cannot be used. We therefore use the ordinary least squares (OLS) estimator with robust standard errors and assuming that observations are independent across countries, but not necessarily within countries over time, i.e. observations are assumed to be clustered. This method is a variant of the Huber-White robust estimator, which provides correct estimation in the presence of any pattern of correlation among errors within units, including serial correlation and correlation due to unit-specific components. In other words, the robust-cluster option produces consistent standard errors even in the presence of serial correlation and heteroskedasticity, but it is potentially inefficient in estimation (Wiggins 1999). We found that standard errors were generally higher using OLS with the robust-cluster option than using the PCSE method without the assumption of contemporaneous correlation, which is why we prefer using the former as this suits our conservative testing approach. To ensure that results are not specific to our estimation technique, we additionally use the Generalized

20 Estimation Equation method (GEE) (Zorn 2001), also under the assumption of clustered observations, and a feasible generalized least squares (GLS) estimator where we allow for panel-heteroskedasticity. Endogeneity is clearly a problem, not for our main variable of interest, but for many of the other explanatory variables. High military expenditures can deter international conflicts, but can also provoke them if foreign countries go for preventive action as Kant (1795) already pointed out more than 200 years ago (Fordham and Walker 2005). High military expenditures can signal to rebels that the initiation of a civil war is likely to end in defeat, but particularly in fragile post-conflict societies high expenditures can also increase the risk of renewed conflict if the former rebels take such expenditures as a signal for the government’s willingness to renege on the peace terms (Collier and Hoeffler 2004c). A high degree of militarization undermines democracy and promotes autocratic tendencies (Kimenyi and Mbaku 1995). The effect of high military expenditures on growth and economic development are not entirely straightforward, but the evidence seems to favour the view that such expenditures are detrimental (Sandler and Hartley 1995). With so many potentially endogenous variables it becomes next to impossible to apply the technique of instrumental variables regression. We therefore resign ourselves to acknowledging the potential endogeneity problems, hoping that the results on our main variable of interest are not adversely affected.

Results Table 3 presents the results for military spending with Fearon (2003) and Fearon and Laitin’s (2003) measures of ethnic and religious heterogeneity. Note that year-specific time dummies are included in the estimations, but their coefficients are not reported to save space. Column 1 reports OLS results with the robust-cluster option, column 2 reports results using the GEE

21 method, and column 3 reports the GLS results. 13 As seen there, ethnic heterogeneity is negatively related to military spending as a share of GDP across all three testing procedures. Religious heterogeneity is not statistically significantly different from zero in any of the estimations. Substantively, holding all other variables at their mean values, raising ethnic heterogeneity from its median to the 95th percentile would reduce the global average share of military expenditures in GDP over this time period by 7 per cent (all computations of substantive results refer to the OLS estimation results). 14 Note that ethnic fractionalization’s substantive effect is computed net of the lagged dependent variable, which tends to dampen the actual effect. The coefficient of the lagged dependent variable also varies somewhat across the estimators used, which has a substantial impact on the size of coefficients of the explanatory variables. The computed substantive effects should therefore be treated with care. Contrary to Goldsmith (2003) who tests a longer time period, we do not find that higher per capita income predicts higher defense spending. This may suggest some influence from the Cold War period that dominates other tests. Developed and Eastern European countries have on average reduced their military spending after the end of the Cold War, whereas developing countries have not, or if they have, by smaller degrees. Democracy has a negative and statistically significant impact on military spending. Our tests support Goldsmith’s (2003) and Collier and Hoeffler’s (2004b) findings on democracy’s effect on lower militarization. Democratic governments are able, independently of the level of fractionalization, wealth, and other controls, to focus a larger share of resources to other

13

Collinearity among the variables does not seem to be a problem. The average Variance Inflation

Factor (VIF) score is around 2 in column 1. 14

We employ the software program CLARIFY to compute the substantive effects (Tomz, Wittenberg,

and King 2003).

22 priorities than security, a result expected by some theorists on regimes and military spending (Russett 1990; Sprout and Sprout 1968). This result is not likely to be driven mainly by the fact that democracies thrive in peaceful neighborhoods and autocrats thrive in violent ones, which can be deduced from the fact that we control for violent conflict. 15 Not surprisingly, larger government consumption in general is also positively related to higher military expenditure. There is of course some partial identity, but dropping the variable from the regression has little effect on the basic results (results available upon request). Military spending by contiguous countries and the incidence of civil and international war all have the expected positive sign and are statistically significant, results that are consistent with others (Goldsmith 2003; Gupta, de Melo, and Sharan 2001). Military expenditures generally do not decrease with a longer history of peace. Only for civil war is the coefficient statistically significant and then only in column 2. We experimented with a simple dummy variable for past conflict experience, but results were not much affected. Either the effect of a longer history of peace is very complex or it does not exist. We dropped the conflict variables to assess the effects of ethnic heterogeneity without them in the model, since these conflicts may in fact be related to the heterogeneity variables. In other words, how sensitive are the results of heterogeneity to the inclusion of conflict itself? When the four conflict variables are dropped, the basic results on the heterogeneity variables change very little, thus heterogeneity’s effect on military spending is not dependent on controlling for the incidence and history of militarized conflicts in the model. We find no evidence that countries with larger population size can enjoy a lower defense burden since they might be less at risk of invasion. Instead, if anything column 3 suggests that larger countries have higher spending. If so, this could be because countries with a large population are likely to

15

There is a large literature on the democratic peace (Russett and Oneal 2000). For neighborhoods and

democracy, see (Gleditsch and Ward 2004; O'Loughlin, Ward, and Lofdahl 1998).

23 have the aspiration of being a major power player and one needs a strong military base to assert such power. To see whether ethnic and religious heterogeneity exerts a non-linear influence on military spending, we repeated the estimations with squared and, in separate estimations, even cubic heterogeneity terms additionally included. However, we found no evidence whatsoever for a non-linear relationship. This holds true for the other dependent variables and the alternative measures of heterogeneity reported below as well. Table 4 reports results for military personnel as a share of the total labour force. As seen there, again, ethnic heterogeneity is negatively related to militarization, even though the effect is only statistically significant in columns 2 and 3. Religious fractionalization still does not matter. A higher growth rate allows governments to engage in greater militarization of the labour force (result not statistically significant in column III). Democracy again shows a statistically significant negative effect on the share of labour devoted to security. Higher government spending and per capita income are positively and population size negatively related to military personnel as a share of the labor force, but only in GLS estimation. Higher militarization by contiguous neighbors leads to higher militarization within the country, but the effect is not statistically significant in column 2. Interestingly, the conflict variables do not seem to exert any influence on this aspect of militarization, except for column 3 where civil war experience is statistically significant with the expected positive sign. This could be an artefact of the way in which statistics are reported when countries militarize temporarily for conflicts. It may also mean that governments engage in armed conflict in a capitalintensive rather than labour-intensive way using military equipment rather than soldiers. Table 5 reports estimation results for the arms imports aspect of militarization. Strikingly, ethnic fractionalization exerts once again a negative influence on militarization, and religious fractionalization fails to matter as before. Higher arms imports by contiguous

24 neighbors and events of civil and international war raise a country’s arms imports. 16 A longer history of peace from international, but not civil, war reduces arms imports in columns 2 and 3. Interestingly, democracies are not estimated to import any lower levels of arms than autocracies. Perhaps surprisingly, arms imports are lower when economic growth is strong, but the effect is not statistically significant in columns 2 and 3. Population size is positively associated with arms imports, but statistically significantly only in column 3. In table 6, we repeat the tests conducted above, but this time using alternative measures of heterogeneity. We report only results from the OLS estimator to save space, but results are largely similar for the other estimation techniques unless otherwise noted. Estimations using Montalvo and Reynal-Querol’s (2005) measures of ethnic and religious fractionalization largely mirror the results using Fearon’s (2003) and Fearon and Laitin’s (2003) measures: More ethnically fractionalised societies have lower militarization for all three dependent variables. Religious fractionalization does not matter for military expenditures and arms imports, but it is positively associated with a higher share of the labor force employed as military personnel (not significant in GEE estimation). Results from the main estimations uphold if either Kheng’s (2001) index of ethnic fractionalization or Roeder’s (2001) ethnolinguistic fractionalization index is used instead (note that neither of the two have constructed a religious fractionalization measure). Like with Fearon’s and Fearon and Laitin’s measures, ethnic fractionalization is not significant in OLS estimation for military personnel, but it is significant in GEE and GLS estimations. Employing Alesina et al.’s (2003) measures of ethnic, linguistic and religious fractionalization suggests that none of the fractionalization measures matter in OLS estimation. However, linguistic fractionalization is close to being statistically significant for military expenditures (p-value 0.107). Also, in

16

The civil war variable is only marginally statistically insignificant in column 1 with a p-value of

0.109.

25 GEE and GLS estimation, linguistic fractionalization becomes marginally significant with a negative coefficient sign for military expenditures and arms imports, but not for the size of the military personnel, whereas ethnic and religious fractionalization remain insignificant throughout. We re-ran all tests by dropping linguistic fractionalization because it is highly correlated with ethnic fractionalization, but the results do not change much. This is not so surprising in that Alesina et al.’s (2003) linguistic rather than ethnic fractionalization measure is most highly correlated with Fearon and Laitin’s ethnic fractionalization measure (r = 0.88 as opposed to r = 0.76). Given that linguistic issues are potentially most explosive, the fact that states do not militarize under linguistic diversity may very well signal that consensus and coordination is most difficult under these conditions. To test this aspect of diversity further, we now test Fearon’s (2003) measure of cultural fractionalization that adjusts his measure of ethnic fractionalization for the cultural distance between the ethnic groups using linguistic classifications of distance between major language families. For example, if ethnic groups belong to two distinct language families, such as Greek and Turkish, then the cultural distance is greater compared to two groups speaking Slavic East branch and Slavic West branch. In other words, the greater the overlap of shared linguistic traits among ethnic groups, the lower the cultural distance between them and the more a country’s index of cultural fractionalization will lie below its index of ethnic fractionalization. Indeed, Fearon (2003: 215) writes that “if a researcher’s theory is that ethnic fractionalization matters because it makes for diverse preferences and consequent difficulties cooperating, then the measure of cultural fractionalization (...) may be more appropriate.” Alesina et al (2003) concur that future research should develop cultural indicators based on the cultural distance concept introduced by Fearon. We find a correlation of r = 0.82 between Fearon’s ethnic and cultural fractionalization measures. We also find that cultural fractionalization is correlated r = 0.74 with Alesina et al’s linguistic and ethnic fractionalization measure. Table 7 repeats the OLS

26 estimations from tables 1 to 3, but replacing ethnic with cultural fractionalization.17 As seen in column 1, cultural fractionalization is also negatively and statistically significantly related to military expenditure as a share of GDP. However, it exerts no statistically significant influence on either the share of military personnel in the total labour force or the ratio of arms to total imports. Religious fractionalization remained insignificant. Finally, we also tested the hypothesis of Montalvo and Reynal-Querol (2005) that it is ethnic and religious polarization rather than fractionalization that matters. To do so, we replaced the fractionalization with their polarization measures (see table 8). As before, we report only results from OLS estimation, but the ones using the other techniques are very similar unless otherwise noted. Polarization has no impact on military expenditures. Ethnic polarization is associated with a smaller size of military personnel, whereas the opposite is the case for religious polarization. However, the result is not robust toward using the GEE estimation technique. Polarization does not seem to matter for arms imports. Only in (nonreported) GLS estimation did ethnic polarization seem to be associated with arms imports, but rather than raising arms imports the estimations suggest that higher polarization leads to lower arms imports. These results basically mirror the ones from using Montalvo and ReynalQuerol’s (2005) measure of fractionalization, which does not suggest that it is polarization that really matters. We ran several additional tests of sensitivity. 18 With our ethnic heterogeneity variables being invariant over time, we cannot compute a fixed effects model. However, we tried to capture some crude cross-regional heterogeneity by regional dummy variables, where the regional classification follows that of World Bank (2004). With the exception of the region of North Africa and Middle East, which often has a higher level of militarization, there

17

Results are qualitatively similar for the other estimation techniques.

18

Results available upon request.

27 was little evidence for systematic regional differences. Our main results were hardly affected with the exception of the militarization of continguous countries, which often became statistically significant. This is no surprise of course given that militarization of contiguous countries is correlated with regional levels of militarization. Next, we limited our analyses to a sub-sample of developing countries. The results on diversity were very similar. The share of the population urbanized had little effect on the results when added to the models, nor was this variable statistically significant contrary to the findings of others (Davoodi et al. 2001). The same is true for the age-dependency ratio, which is supposed to drive higher welfare spending, thus potentially lowering military spending (Davoodi et al. 2001; Gupta, de Melo, and Sharan 2001) and for the level of aid to gross national income, which might ease the governmental budget constraint. One might wonder whether oil wealth might allow governments greater levels of militarization. Adding a dummy variable taking the value of one if oil exports reach one third of total GDP (Fearon and Laitin 2003), suggests no impact on military expenditures or military personnel, but has a positive and statistically significant effect on arms imports. This concurs with the fact that major oil exporters such as the Gulf countries have been major arms importers over the last decade or so. The results on the remaining variables were hardly affected, however. The same is true if we add a dummy variable for the 20 largest arms-producing countries based on information from SIPRI (2002) to the arms imports regressions. Major arms producers import fewer arms, as expected, but our results remain valid. Next, we added a measure of institutional quality (the ICRG index) now available from World Bank (2004). This variable was statistically insignificant and did not change the basic results reported above, despite a significant drop in sample size. The same is true if we use the sub-component of the ICRG index that measures corruption. Interacting institutional quality with diversity does not suggest that ethnicity’s effect is

28 contingent on institutional quality either.19 It seems that ethnicity’s effects are not sensitive to the quality of institutions as suggested by Easterly (2001). Future research could pursue this issue further. In sum, there is no indication from any of the tests that fractionalization increases militarization. The same is true for polarization. The only possible exceptions are religious fractionalization as measured by Montalvo and Reynal-Querol (2005) and religious polarization, for which there is some weak evidence that it might increase the size of military personnel in one two of three estimations (OLS and GEE as opposed to GLS), without having an effect on the other two measures of militarization, however. These results disconfirm what we would expect to see from theoretical arguments about ethnic and other diversity, fear, and militarization.

Conclusions Several disciplines use ethnic diversity as an explanation for societal outcomes ranging from democratization and governance issues to violent political conflict and economic performance. Recent empirical studies, particularly among economists, show that ethnic diversity hampers public goods provision because consensus and coordination are difficult to achieve under conditions of diverse preferences. In fact, a major implication from this literature would be to look favorably on an enlightened dictator, who simply allocates public goods based on welfare maximizing criteria, rather than allowing diverse preferences to result in bad policy outcomes. Such views are heavily contested, of course (Collier 2001b). In addition, explanations of violent conflict see ethnic diversity and cultural difference as problematic because it can lead to mutual hatred and fear stemming from historic legacies, and may lead to security dilemmas and domination and control by majority-controlled states

19

An interaction term between democracy and diversity was similarly insignificant throughout.

29 of minority groups and vice versa. Cross-national quantitative studies do not find that ethnic diversity spurs violent conflict, however. The question our study was concerned with is: are the empirical results on the link between diversity and public goods provision, such as education and infrastructure, driven by the fact that governments deal with diversity by spending more on militarization, to the extent that it crowds out public goods not related to the security sector? After all, security is a public good, and if some of the most pessimistic theories of ethnic diversity and ethnic conflict are true, then states might be responding first and foremost with militarization to security considerations believing that higher militarization leads to greater security. Our results simply do not support this view. Militarization is actually lower under conditions of greater diversity measured by several different indicators. Ethno-linguistic diversity in particular seems to be what matters rather than polarization. If in fact, as some find, ethnicized armed violence is most likely when two groups are of similar size, then it is not likely that military spending through which states may increase their hold over society is the link as to why highly heterogeneous societies are better able to maintain peace (Collier and Hoeffler 2004a; Garcia-Montalvo and Reynal-Querol 2002). Clearly, wider coverage of militarization variables beyond spending, the size of military forces and arms imports will provide more opportunities of testing the effects of ethnic and other diversity on states’ response to real and perceived threats. Perhaps empirical studies in the future could include those expenditures classified as internal security expenditure on police and paramilitary forces, both public and private. Certainly, more research should be conducted to address the issue of how in fact ethnicity, market outcomes, and reduction of possible conflict may be affecting state behavior in the broader realm of securitization. On the other hand, further research should unpack the messy connections between ethnic diversity and the politics of patronage that erode state provision of public goods. If ethnic politics leads to under-provision of public goods, then the public good of security

30 might also be affected in a similar manner as other public goods, perhaps with deadly consequences. It remains to be seen, however, as to whether or not the lower militarization in more diverse societies is a peace liability, or a peace dividend. Collier and Hoeffler (2004b) do not find that higher military spending deters civil conflict, whilst Collier and Hoeffler (2004c) even show that higher spending might increase rather than reduce the risk of renewed conflict in fragile post-conflict societies. If so, then lower militarization might represent a blessing rather than a problem. Security is a public good, but militarization might not provide it. Further research should try to establish if, rather than through militarization, security is bought through corruption and patronage in ethnically diverse societies instead. If so, then the World Bank and other international agencies that push transparency and the fight against corruption may in fact increase vulnerabilities. But without further research, this remains speculation. Fearon and Laitin (1996) developed theories of interethnic cooperation built on in-group policing and fear of spirals of conflict, which might provide better explanations for the fact that highly fractionalized societies are surprisingly peaceful. Our analyses suggest clearly that ethnic diversity promotes lower militarization, which means that diversity promises substantial dividends if high levels of militarization is in fact undesirable. Indicative of the latter is that democracy in general promotes increased public goods provision (Baum and Lake 2003; Boix 2001; Lake and Baum 2001), but leads to lower rather than higher militarization. Promoting democracy and diversity, not dictatorship and secession, might therefore provide the best option for promoting increased public goods provision with simultaneously low levels of militarization.

31 References: Achen, Christopher. 2000. Why Lagged Dependent Variables Can Suppress the Explanatory Power of Other Independent Variables. Paper read at Annual Meeting of the Political Methodology Section of the American Political Science Association, UCLA, July 2022, 2000. Alesina, Alberto. 1994. Political Models of Macroeconomic Policy and Fiscal Reforms. In Voting for Reform: Democracy, Political Liberalization, and Economic Reform, edited by S. Haggard and S. Webb. Oxford: Oxford University Press. Alesina, Alberto, Reza Baqir, and William Easterly. 1999. Public Goods and Ethnic Divisions. Quarterly Journal of Economics 114 (4):1243–1284. Alesina, Alberto, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and Romain Wacziarg. 2003. Fractionalization. Journal of Economic Growth 8:155–194. Alesina, Alberto, and Allan Drazen. 1991. Why Are Stabilizations Delayed? American Economic Review 81 (5):1170–1188. Alesina, Alberto, and Dani Rodrik. 1994. Distributive Politics and Economic Growth. Quarterly Journal of Economics 109 (2):465-490. Baum, Matthew A., and David A. Lake. 2003. The Political Economy of Growth: Democracy and Human Capital. American Journal of Political Science 47 (2):333–347. Beck, Nathaniel, and Jonathan N. Katz. 1995a. Nuisance vs. Substance: Specifying and Estimating Time-Series–Cross-Section Models. Political Analysis 6. Beck, Nathaniel, and Jonathan N. Katz. 1995b. What To Do (and Not To Do) with TimeSeries Cross-Section Data. American Political Science Review 89 (3):634-647. Bennet, D. Scott, and Allen C.. Stam. EUGene v3.0. 2003 [cited. Available from http://www.eugenesoftware.org. Boix, Carles. 2001. Democracy, Development, and the Public Sector. American Journal of Political Science 45 (1):1–17. Brzoska, Michael. 1995. World Military Expenditures. In Handbook of Defense Economics, edited by K. Hartley and T. Sandler. Oxford: Elsevier. Collier, Paul. 2001a. Ethnic Diversity: An Economic Analysis of Its Implications. Economic Policy 32:129–166. Collier, Paul. 2001b. Implications of Ethnic Diversity. Washington, DC: World Bank. Collier, Paul , and Anke Hoeffler. 2004a. Greed and Grievance in Civil War. Oxford Economic Papers 56 (4):563–595. Collier, Paul, and Anke Hoeffler. 2002. Military Expenditure: Threats, Aid and Arms Races. Washington, DC: World Bank. Collier, Paul, and Anke Hoeffler. 2004b. Military Expenditure in Post-conflict Societies. Oxford: Center for the Study of African Economies. Collier, Paul, and Anke Hoeffler. 2004c. Military Expenditure: Threats, Aid and Arms Races. Oxford: Center for the Study of African Economies. Davoodi, Hamid, Benedict Clements, Jerald Schiff, and Peter Debaere. 2001. Military Spending, the Peace Dividend, and Fiscal Adjustment. IMF Staff Papers 48 (2):290– 316. de Soysa, Indra. 2002. Paradise is a Bazaar? Greed, Creed, and Governance in Civil War, 1989–1999. Journal of Peace Research 39 (4):395–416. Drake, St. Claire. 1957. Some Observations on Interethnic Conflicts as One Type of Intergroup Conflict. Journal of Conflict Resolution 1 (2):155–178. Easterly, William. 2001. Can Institutions Resolve Ethnic Conflict? Economic Development and Cultural Change 49 (4):687–706. Easterly, William, and Ross Levine. 1997. Africa' Growth Tragedy: Policies and Ethnic Divisions. Washington D.C.: World Bank.

32 Ellingsen, Tanja. 2000. Colorful Community or Ethnic Witches' Brew?: Multiethnicity and Domestic Conflict During and After the Cold War. Journal of Conflict Resolution 44 (2):228–249. Fearon, James D. 2003. Ethnic and Cultural Diversity by Country. Journal of Economic Growth 8:195–222. Fearon, James D., and David D. Laitin. 2003. Ethnicity, Insurgency, and Civil War. American Political Science Review 97 (1):1–16. Fordham, Benjamin O., and Thomas C. Walker. 2005. Kantian Liberalism, Regime Type, and Military Resource Allocation: Do Democracies Spend Less? International Studies Quarterly 49 (1):141–157. Fox, Jonathan. 2004. The Rise of Religious Nationalism and Conflict: Ethnic Conflict and Revolutionary Wars, 1945–2001. Journal of Peace Research 41 (6):715–731. Garcia-Montalvo, José, and Marta Reynal-Querol. 2002. Why Ethnic Fractionalization? Polarization, Ethnic Conflict, and Growth. Barcelona: Universitat Pompeu Fabra. Garcia-Montalvo, José, and Marta Reynal-Querol. 2005. Ethnic Diversity and Economic Development. Journal of Development Economics 76 (1):293–323. Gilley, Bruce. 2004. Against the Concept of Ethnic Conflict. Third World Quarterly 25 (6):1155–1166. Gleditsch, Kristian, and Michael Ward. 2004. Diffusion and the International Context of Democratization. La Jolla, CA: University of California, San Diego. Gleditsch, Nils Petter, Göran Lindgren, Naima Mouhleb, Sjoerd Smit, and Indra de Soysa. 2000. Making Peace Pay: A Bibliography on Disarmament and Conversion. Claremont, CA: Regina. Gleditsch, Nils Petter, Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg, and Havard Strand. 2002. Armed Conflict 1946–2001: A New Dataset. Journal of Peace Research 39 (5):615–637. Goldsmith, Benjamin E. 2003. Bearing the Defence Burden, 1886–1989: Why Spend More? Journal of Conflict Resolution 47 (5):551–573. Gupta, Sanjeev, Luiz de Melo, and Raju Sharan. 2001. Corruption and Military Spending. European Journal of Political Economy 17:749–777. Gurr, Ted R. 1970. Why Men Rebel. Princeton, NJ: Princeton University Press. Gurr, Ted R. 1993. Minorities at Risk: A Global View of Ethnopolitical Conflict. Washington, DC: United States Institute of Peace Press. Harrison, Lawrence, and Samuel P. Huntington, eds. 2000. Culture Matters How Values Shape Human Progress. New York: Basic Books. Hartley, Keith, and Nicholas Hooper. 1990. The Economics of Defense, Disarmament, and Peace: An Annotated Bibliography. Aldershot: Elgar. Henderson, Errol A. 1997. Culture or Contiguity: Ethnic Conflict, the Similarity of States, and the Onset of War, 1820–1989. Journal of Conflict Resolution 41 (5):649–668. Horowitz, Donald L. 1998. Structure and Strategy in Ethnic Conflict: A Few Steps Towards Synthesis. Paper read at Annual Bank Conference on Development Economics, at Washington, DC. Huntington, Samuel P. 1968. Political Order in Changing Societies. New Haven, CT: Yale University Press. Huntington, Samuel P. 1993. The Clash of Civilizations. Foreign Affairs 72 (3):22–49. Kaufmann, Chaim. 1996. Possible and Impossible Solutions to Ethnic Civil Wars. International Security 20 (4):136–175. Keefer, Philip, and Stephen Knack. 2002. Polarization, Politics and Property Rights: Links between Inequality and Growth. Public Choice 111:127–154. Kheng, Yeoh Kok. 2001. Towards an Index of Ethnic Fractionalization. Kuala Lumpur, Malaysia: University of Malaya.

33 Kimenyi, Mwangi S. , and Mukum John Mbaku. 1995. Rents, Military Elites, and Political Democracy. European Journal of Political Economy 11:699–708. La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Schleifer, and Robert Vishny. 1998. The Quality of Government. Vol. NBER Working paper # 6727. Cambridge, MA: NBER. Lake, David A., and Matthew A. Baum. 2001. The Invisible Hand of Democracy: Political Control and the Provision of Public Services. Comparative Political Studies 34 (6):587–621. Maizels, Alfred, and Machiko K. Nissanke. 1986. The Determinants of Military Expenditures in Developing Countries. World Development 14 (9):1125–1140. Marshall, Monty, and Ted R. Gurr. 2003. Peace and Conflict 2003. College Park, MD: CIDCM, University of Maryland. Mearsheimer, John J. 1990. Back to the Future: Instability in Europe After the Cold War. International Security 15 (1):5–56. Mueller, John. 2000. The Banality of Ethnic War. International Security 25 (1):42–70. O'Loughlin, Michael Ward, and Corey L. Lofdahl. 1998. The Diffusion of Democracy, 1946– 1994. Annals of the Association of American Geographers 88 (4):545–574. Olson, Mancur. 1982. The Rise and Decline of Nations: Economic Growth, Stagflation, and Social Rigidities. New Haven: Yale University Press. Omitoogun, Wuyi. 2003. Military Expenditure Data in Africa: A Survey of Cameroon, Ethiopia, Ghana, Kenya, Nigeria, and Uganda, SIPRI Research Report. Oxford: Oxford University Press, for SIPRI. Petersen, Roger. 2001. Understanding Ethnic Violence: Fear, Hatred, Resentment in Twentieth Century Eastern Europe. Cambridge: Cambridge University Press. Posen, Barry. 1993. The Security Dilemma and Ethnic Conflict. Survival 35:27–47. Posner, Daniel N. 2004. Measuring Ethnic Fractionalization in Africa. American Journal of Political Science 48 (4):849–863. Ramet, Sabrina. 2004. 'For a Charm of Pow'rful Trouble, Like a Hell-broth Boil and Bubble': Theories About the Roots of the Yugoslav Troubles. Nationalities Papers 32 (4):731– 763. Reynal-Querol, Marta. 2002. Ethnicity, Political Systems, and Civil Wars. Journal of Conflict Resolution 46 (1):29–54. Roe, Paul. 2000. Former Yugoslavia: The Security Dilemma that Never Was? European Journal of International Relations 6 (3):373–393. Roeder, Phillip G. 2001. Ethnolinguistic Fractionalization Indices, 1961 and 1985. La Jolla, CA: University of California, San Diego. Russett, Bruce. 1990. Controlling the Sword: The Democratic Governance of National Security. Cambridge, MA: Harvard University Press. Russett, Bruce, and John Oneal. 2000. Triangulating Peace: Democracy, Interdependence, and International Organizations, The Norton Series in World Politics. London: W. W. Norton and Company. Russett, Bruce, John R. Oneal, and Michealene Cox. 2000. Clash of Civilizations, or Realism and Liberalism Déjà Vu? Some Evidence. Journal of Peace Research 37 (5):583–608. Sambanis, Nicholas. 2001. Do Ethnic and Nonethnic Civil Wars Have the Same Causes? A Theoretical and Empirical Inquiry (Part 1). Journal of Conflict Resolution 45 (3):259– 282. Sandler, Todd, and Keith Hartley. 1995. The Economics of Defense. Cambridge: Cambridge University Press. Snyder, Jack, and Robert Jervis. 1999. Civil War and the Security Dilemma. In Civil Wars, Insecurity, and Intervention, edited by B. F. Walter and J. Snyder. New York: Columbia University Press.

34 Sprout, Harold, and Margaret Sprout. 1968. The Dilemma of Rising Demands and Insufficient Resources. World Politics 20 (4):660–693. Tomz, Michael, Jason Wittenberg, and Gary King. 2003. Clarify: Software for Interpreting and Presenting Statistical Results. Stanford: Stanford University, Department of Political Science. Varshney, Ashutosh. 2001. Ethnic Conflict and Civil Society: India and Beyond. World Politics 53:362–398. Walter, Barbara F., and Jack Snyder, eds. 1999. Civil Wars, Insecurity, and Intervention. New York: Columbia University Press. Wiggins, Vince. Comparing XTGLS with Regress Cluster () Stata Corporation, 1999 [cited. Available from www.stata.com/support/faqs/stgls_rob.html. Wimmer, Andreas, Richard J. Goldstone, Donald L. Horowitz, Ulrike Joras, and Conrad Schetter, eds. 2004. Facing Ethnic Conflicts: Towards a New Realism. Oxford: Rowman and Littlefield. World Bank. 2003. World Development Indicators CD Rom. Washington, DC: The World Bank. World Bank. 2004. World Development Indicators CD Rom. Washington, DC: World Bank. Zorn, Christopher. 2001. Generalized Estimation Equation Models for Correlated Data: A Review with Applications. American Journal of Political Science 45 (2):470–490. Kant, Immanuel. 1795 [1991]. Perpetual Peace. In Kant’s Political Writings, 2nd edition, translated by H.B. Nisbert and H. Reiss. Cambridge: Cambridge University Press. Fearon, James D. and David D. Laitin. 1996. Explaining Interethnic Cooperation. American Political Science Review 90 (4):715-735. SIPRI. 2002. Military Expenditure and Arms Production Project. Stockholm: Stockholm International Peace Research Institute. www.sipri.org/contents/milap/milex/aprod/nationaldata/summary.pdf

Table 1. Correlation matrix of fractionalization and polarization measures.

1: ethfrac (Fearon & Laitin) 2: relfrac (Fearon & Laitin) 3: cultfrac (Fearon) 4: ethfrac (Alesina et al.) 5: relfrac (Alesina et al.) 6: linfrac (Alesina et al.) 7: ethlinfrac (Roeder) 8: ethfrac (Kheng) 9: ethfrac (Montalvo & Reynal-Querol) 10: relfrac (Montalvo & Reynal-Querol) 11: ethpol (Montalvo & Reynal-Querol) 12: relpol (Montalvo & Reynal-Querol)

1 1.00 0.39 0.82 0.76 0.31 0.88 0.85 0.84 0.84 0.51 0.42 0.54

2

3

4

5

6

7

8

9

10

11

1.00 0.33 0.31 0.89 0.40 0.43 0.30 0.36 0.50 0.12 0.46

1.00 0.74 0.20 0.74 0.70 0.76 0.68 0.50 0.39 0.51

1.00 0.23 0.68 0.83 0.89 0.81 0.53 0.55 0.59

1.00 0.31 0.36 0.24 0.29 0.51 0.11 0.47

1.00 0.76 0.76 0.73 0.48 0.28 0.50

1.00 0.91 0.86 0.52 0.51 0.58

1.00 0.87 0.56 0.57 0.63

1.00 0.54 0.58 0.57

1.00 0.32 0.96

1.00 0.40

Table 2. Summary descriptive variable information. Variable military expenditures per GDP military personnel per labor force arms imports to total imports ethfrac (Fearon & Laitin) relfrac (Fearon & Laitin) cultfrac (Fearon) ethfrac (Alesina et al.) relfrac (Alesina et al.) linfrac (Alesina et al.) ethlinfrac (Roeder) ethfrac (Kheng) ethfrac (Montalvo & Reynal-Querol) relfrac (Montalvo & Reynal-Querol) ethpol (Montalvo & Reynal-Querol) relpol (Montalvo & Reynal-Querol) ln GNI p.c. economic growth polity government expenditures ln population contiguous military expenditures contiguous military personnnel contiguous arms imports peace years (civil war) peace years (international conflict) civil war international conflict

Obs 1442 1178 1186 1441 1441 1418 1442 1442 1415 1442 1442 1185 1198 1185 1198 1442 1442 1442 1442 1442 1442 1178 1186 1442 1442 1442 1442

Mean 2.91 1.50 2.48 0.40 0.36 0.29 0.43 0.42 0.39 0.46 0.46 0.44 0.28 0.51 0.46 8.36 0.04 3.08 16.23 16.33 2.97 1.28 3.02 20.13 26.93 0.19 0.02

Std. Dev. 2.81 1.82 6.66 0.28 0.21 0.21 0.25 0.23 0.28 0.27 0.26 0.28 0.23 0.24 0.34 1.15 0.05 6.87 6.70 1.47 2.26 1.20 5.61 19.03 17.62 0.39 0.14

Min 0.30 0.08 0 0.00 0 0 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.02 0.00 5.97 -0.35 -10 2.98 12.90 0.45 0.09 0 0 0 0 0

Max 29 23.68 103.30 0.93 0.78 0.73 0.93 0.86 0.92 0.98 0.88 0.96 0.78 0.98 1 10.49 0.25 10 56.51 20.97 24.46 9.43 36.86 56 56 1 1

37 Table 3. Military spending and Fearon and Laitin’s measures of fractionalization, 1988–2002

military expenditures (lagged) ethfrac relfrac ln GNI p.c. economic growth polity government expenditures ln population contiguous military spending peace years (civil war) peace years (intern. conflict) civil war international conflict Observations R-squared Countries

(1) OLS 0.783*** (14.18) -0.414** (2.29) 0.144 (0.95) -0.029 (0.43) -1.066 (1.22) -0.020** (2.35) 0.039* (1.95) 0.035 (1.23) 0.065** (2.15) -0.003 (1.36) 0.001 (0.59) 0.259*** (3.33) 1.377* (1.85) 1455 0.87

(2) GEE 0.410*** (5.14) -1.316*** (3.10) 0.435 (1.04) -0.174 (1.00) -0.100 (0.15) -0.032** (2.37) 0.103** (2.54) 0.053 (0.87) 0.106** (2.17) -0.009*** (2.83) -0.001 (0.23) 0.393*** (3.61) 1.136* (1.84) 1455

(3) GLS 0.894*** (92.76) -0.109** (2.40) 0.061 (1.26) 0.012 (0.84) -0.659*** (2.91) -0.007*** (3.35) 0.009*** (3.99) 0.012* (1.73) 0.019*** (2.60) -0.001 (1.24) -0.000 (0.36) 0.135*** (4.58) 0.208** (2.34) 1455

129

129

Notes: t- and z-statistics in brackets. Constant and year-specific time-dummies included, but coefficients not reported. *, **, *** significant at .1, .05 and .01 level, respectively.

38 Table 4. Military personnel and Fearon and Laitin’s measures of fractionalization, 1988–2002

military personnel (lagged) ethfrac relfrac ln GNI p.c. economic growth polity government expenditures ln population contiguous military personnel peace years (civil war) peace years (intern. conflict) civil war international conflict Observations R-squared Countries

(1) OLS 0.845*** (14.27) -0.105 (1.44) -0.007 (0.11) 0.041 (1.11) 0.547** (2.32) -0.009* (1.89) 0.007 (1.08) -0.016 (1.39) 0.036*

(2) GEE 0.533*** (4.17) -0.606* (1.94) -0.284 (1.04) 0.025 (0.30) 0.490** (2.15) -0.007* (1.92) 0.018 (1.41) -0.053 (1.30) 0.022

(3) GLS 0.928*** (132.10) -0.027** (2.13) -0.020 (1.26) 0.010*** (2.60) 0.014 (0.22) -0.002*** (3.23) 0.001 (1.48) -0.006*** (2.87) 0.015**

(1.79) -0.001 (1.29) -0.000

(0.31) -0.002 (1.07) -0.002

(2.55) -0.000 (0.83) 0.000

(0.49) 0.054 (1.22) 0.331 (1.11) 1255 0.94

(0.48) 0.029 (0.85) 0.337 (1.03) 1255

(0.32) 0.018** (1.98) 0.010 (0.32) 1255

137

137

Notes: t- and z-statistics in brackets. Constant and year-specific time-dummies included, but coefficients not reported. *, **, *** significant at .1, .05 and .01 level, respectively.

39 Table 5. Arms imports and Fearon and Laitin’s measures of fractionalization, 1988–2002

arms imports (lagged) ethfrac relfrac ln GNI p.c. economic growth polity government expenditures ln population contiguous arms imports peace years (civil war) peace years (intern. conflict) civil war international conflict Observations R-squared Countries

(1) OLS 0.521*** (5.74) -1.266*** (2.87) -0.074 (0.13) 0.040 (0.20) -3.421* (1.66) -0.047 (1.24) 0.035 (1.01) 0.088 (1.10) 0.086** (1.98) -0.006 (0.94) -0.009 (1.25) 0.808* (1.74) 2.223* (1.91) 1271 0.58

(2) GEE 0.397*** (7.08) -1.878*** (2.99) 0.026 (0.03) -0.250 (0.95) -2.980 (1.53) -0.018 (0.44) 0.056 (1.31) 0.126 (1.26) 0.105** (2.27) -0.008 (1.29) -0.016** (1.99) 0.998** (2.48) 2.156* (1.82) 1271

(3) GLS 0.554*** (27.49) -0.277** (2.41) 0.042 (0.36) 0.040 (1.31) -0.687 (1.19) -0.004 (0.74) -0.002 (0.40) 0.038** (2.25) 0.045*** (3.28) -0.001 (0.84) -0.003* (1.72) 0.298*** (3.06) 0.360* (1.80) 1271

138

138

Notes: t- and z-statistics in brackets. Constant and year-specific time-dummies included, but coefficients not reported. *, **, *** significant at .1, .05 and .01 level, respectively.

Table 6. Militarization and alternative measures of fractionalization (OLS estimates). Lagged dep. variable ethfrac (Montalvo & R-Q.) relfrac (Montalvo & R-Q.)

(1) mil. exp. 0.792*** (11.68) -0.371** (2.01) 0.198 (1.02)

(2) mil. pers. 0.839*** (14.48) -0.155** (2.10) 0.187** (2.25)

(3) arms imp. 0.531*** (5.45) -1.336** (2.60) 0.026 (0.04)

ethfrac (Kheng)

(4) mil. exp. 0.788*** (14.77)

(5) mil. pers. 0.847*** (14.61)

(6) arms imp. 0.522*** (5.67)

-0.367** (2.04)

-0.055 (0.80)

-1.189** (2.48)

ethlinfrac (Roeader)

(7) mil. exp. 0.788*** (14.94)

(8) mil. pers. 0.846*** (14.63)

(9) arms imp. 0.522*** (5.61)

-0.352** (2.17)

-0.117 (1.40)

-1.021** (2.33)

ethfrac (Alesina et al.) relfrac (Alesina et al.) linfrac (Alesina et al.)

Ln GNI p.c. economic growth polity government expenditures Ln population contiguous militarization peace years (civil war) peace years (intern. confl.) civil war international conflict Observations R-squared

0.019 (0.36) -1.402 (0.96) -0.017* (1.73) 0.026 (1.65) 0.015 (0.60) 0.054* (1.84) -0.003 (1.11) -0.001 (0.44) 0.231*** (2.99) 0.646 (1.55) 1199 0.88

0.084*** (2.75) 0.357 (1.30) -0.008* (1.67) 0.001 (0.20) -0.020* (1.77) 0.038* (1.86) -0.001 (1.21) -0.001 (1.05) 0.073 (1.65) 0.016 (0.25) 1054 0.95

0.180 (1.07) -6.635** (2.27) -0.063 (1.42) 0.028 (0.74) 0.102 (1.18) 0.066 (1.42) -0.010 (1.20) -0.012 (1.57) 0.655 (1.33) 0.808 (1.28) 1059 0.64

-0.031 (0.45) -1.065 (1.20) -0.019** (2.30) 0.038* (1.92) 0.034 (1.26) 0.070** (2.26) -0.003 (1.30) 0.001 (0.74) 0.259*** (3.34) 1.357* (1.84) 1456 0.87

0.046 (1.18) 0.452** (2.10) -0.008* (1.87) 0.006 (1.01) -0.014 (1.32) 0.042** (2.05) -0.001 (1.23) -0.000 (0.37) 0.052 (1.19) 0.331 (1.11) 1264 0.94

0.064 (0.29) -3.500* (1.84) -0.043 (1.14) 0.024 (0.72) 0.095 (1.19) 0.105** (2.39) -0.005 (0.80) -0.008 (1.07) 0.810* (1.76) 2.218* (1.89) 1280 0.57

-0.028 (0.42) -1.100 (1.23) -0.018** (2.22) 0.039* (1.96) 0.035 (1.28) 0.066** (2.17) -0.003 (1.28) 0.001 (0.66) 0.266*** (3.37) 1.350* (1.83) 1456 0.87

0.041 (1.15) 0.435** (2.11) -0.008* (1.87) 0.007 (1.06) -0.013 (1.27) 0.038* (1.85) -0.001 (1.35) -0.000 (0.38) 0.053 (1.22) 0.327 (1.09) 1264 0.94

0.082 (0.34) -3.509* (1.84) -0.039 (1.03) 0.027 (0.81) 0.096 (1.24) 0.101** (2.31) -0.005 (0.81) -0.008 (1.10) 0.817* (1.77) 2.177* (1.86) 1280 0.57

(10) mil exp. 0.788*** (14.48)

(11) mil.pers. 0.847*** (14.14)

(12) arms imp. 0.523*** (5.73)

-0.028 (0.16) -0.074 (0.59) -0.260 (1.63) -0.021 (0.30) -1.284 (1.39) -0.019** (2.30) 0.039* (1.91) 0.033 (1.17) 0.066** (2.16) -0.003 (1.21) 0.001 (0.49) 0.252*** (2.98) 1.391* (1.88) 1429 0.87

0.009 (0.10) -0.049 (0.69) -0.008 (0.11) 0.050 (1.23) 0.453** (2.03) -0.008* (1.72) 0.007 (0.98) -0.013 (1.26) 0.040* (1.96) -0.001 (1.15) -0.000 (0.41) 0.036 (0.85) 0.337 (1.12) 1234 0.94

-0.497 (0.84) -0.280 (0.47) -0.590 (1.26) 0.050 (0.24) -2.458 (1.55) -0.044 (1.14) 0.028 (0.78) 0.097 (1.18) 0.102** (2.26) -0.003 (0.56) -0.009 (1.27) 0.740 (1.44) 2.158* (1.83) 1250 0.58

Notes: t-statistics in brackets. Constant and year-specific time-dummies included, but coefficients not reported. *, **, *** significant at .1, .05 and .01 level, respectively.

Table 7. Militarization and ethnic fractionalization adjusted for cultural/linguistic distance (OLS estimates).

Lagged dependent variable cultfrac (Fearon) relfrac (Fearon & Laitin) ln GNI p.c. economic growth polity government expenditures ln population contiguous militarization peace years (civil war) peace years (intern. confl.) civil war international conflict Observations R-squared

(1) military expenditures 0.787*** (14.56) -0.357* (1.69) 0.110 (0.71) -0.000 (0.01) -1.044 (1.17) -0.020** (2.28) 0.037* (1.91) 0.027 (1.06) 0.070** (2.24) -0.002 (1.15) 0.001 (0.62) 0.265*** (3.33) 1.396* (1.84) 1432 0.87

(2) military personnel 0.846*** (14.35) 0.005 (0.08) -0.039 (0.51) 0.050 (1.27) 0.566** (2.32) -0.008* (1.79) 0.007 (1.06) -0.017 (1.52) 0.039* (1.88) -0.001 (0.97) -0.000 (0.57) 0.055 (1.23) 0.335 (1.10) 1237 0.94

(3) arms imports 0.527*** (5.61) -0.505 (0.68) -0.380 (0.52) 0.144 (0.62) -3.480* (1.67) -0.043 (1.16) 0.031 (0.92) 0.068 (0.89) 0.093** (2.02) -0.003 (0.54) -0.009 (1.24) 0.825* (1.69) 2.266* (1.89) 1251 0.57

Notes: t-statistics in brackets. Constant and year-specific time-dummies included, but coefficients not reported. *, **, *** significant at .1, .05 and .01 level, respectively.

42 Table 8. Militarization and ethnic and religious polarization (OLS estimates).

Lagged dependent variable ethpol (Montalvo & R-Q.) relpol (Montalvo & R-Q.) ln GNI p.c. economic growth polity government expenditures ln population contiguous militarization peace years (civil war) peace years (intern. confl.) civil war international conflict Observations R-squared

(1) military expenditures 0.797*** (11.99) -0.089 (0.56) 0.042 (0.32) 0.053 (0.94) -1.327 (0.91) -0.017* (1.77) 0.024 (1.57) 0.002 (0.11) 0.053* (1.82) -0.003 (0.99) -0.001 (0.33) 0.218*** (2.69) 0.636 (1.54) 1199 0.88

(2) military personnel 0.843*** (14.86) -0.131** (2.16) 0.111** (2.11) 0.102*** (2.98) 0.393 (1.40) -0.008* (1.71) 0.000 (0.01) -0.024** (2.06) 0.037* (1.81) -0.001 (1.51) -0.001 (0.90) 0.067 (1.56) 0.009 (0.14) 1054 0.95

(3) arms imports 0.533*** (5.52) -1.208 (1.54) -0.019 (0.04) 0.331* (1.79) -6.556** (2.23) -0.059 (1.43) 0.025 (0.66) 0.049 (0.64) 0.073* (1.69) -0.013 (1.34) -0.011 (1.33) 0.615 (1.25) 0.792 (1.23) 1059 0.64

Notes: t-statistics in brackets. Constant and year-specific time-dummies included, but coefficients not reported. *, **, *** significant at .1, .05 and .01 level, respectively.