Danger all Along How foreign military threats and domestic power-sharing shape autocratic state-building 20 22 Working paper series 2022:7 Magnus Åsblad Danger all Along How foreign military threats and domestic power-sharing shape autocratic state-building Magnus Åsblad WORKING PAPER SERIES 2022:7 QoG THE QUALITY OF GOVERNMENT INSTITUTE Department of Political Science University of Gothenburg Box 711, SE 405 30 GÖTEBORG December 2022 ISSN 1653-8919 © 2022 by Magnus Åsblad. All rights reserved. Danger all Along. How foreign military threats and domestic power-sharing shape autocratic state-building. Magnus Åsblad QoG Working Paper Series 2022:7 December 2022 ISSN 1653-8919 Abstract ** Standard international relations accounts hold that military buildup by neighboring states leads to increased state-building efforts by authoritarian rulers. In contrast, this study shows that only some dictators will respond by building up stronger states. The reason is that autocrats worry not only about foreign threats but also about threats from within. Power-sharing institutions reduce intra-elite conflict, making such threats less pertinent. Dictatorships with power-sharing institutions are thus less concerned with internal threats and can respond to foreign threats with more ambitious state-building efforts. Using TSCS-data for the world’s autocracies between 1960-2009, the study finds strong support for these claims. Magnus Åsblad The Quality of Government Institute Department of Political Science University of Gothenburg magnus.asblad@gu.se ** Supplementary material for this article is available in the appendix. 2 How do authoritarian rulers react when neighboring states start to ramp-up their militaries? Dominant theories of international politics suggest that such developments often risk triggering an arms race. Uncer- tainty about neighboring states’ intentions may lead governments to respond by increasing their own mili- tary capabilities (Jervis, 1978; Richardson, 1960; Yesilyurt & Elhorst, 2017). The effect of external threats, however, sometimes goes beyond rearmament. A number of studies indicate that the threat of inter-state military conflict has a positive effect on state-building (Gibler & Miller, 2014; Thies, 2007; Thies, 2004). Mainly focusing on the postwar era, these studies draw inspiration from canonical work attempting to explain the rise of strong, capable states in Europe during the early modern period (Downing, 1992; Tilly, 1992; Tilly et al., 1975). In that context, military threats from neighboring states pushed rulers towards state-building – be it in the form of increased taxation, the creation of professional bureaucracies, or the introduction of censuses. Early modern Europe, like many contemporary states, consisted of polities that we today would character- ize as autocracies. And autocrats, this study contends, have an urgent need to worry about power challenges from domestic elites in addition to threats from abroad. All dictators rely on a wider group of regime insiders to stay in power, and the interaction between the ruler and this wider elite group has huge implica- tions for the life of dictatorships (Geddes et al., 2018; Svolik, 2012). In this article, I argue that this rela- tionship also affects the extent to which autocracies respond to external military threats with state-building. Autocrats who continuously fear being deposed by other regime insiders have little time and energy to create a stronger state, despite military buildup at their borders. Even if domestic elites cannot credibly threaten to overthrow the dictator, their opposition may nevertheless make state-building attempts futile. Moreover, an autocratic regime plagued by fear and mistrust tends to reward loyalty over competence when appointing top bureaucrats, thus undermining the creation of an effective state apparatus. In making these claims, the article makes a number of crucial contributions to scholarship on state-building at the intersection of international relations and comparative politics, as will be developed in the following section. Firstly, in contrast to most existing research on state power in autocracies, this study treats state capacity as an outcome to be explained rather than as an independent or moderating variable. Secondly, the article utilizes a new operationalization of external threats that addresses many of the problems associated with existing measures. Thirdly, and most importantly, by incorporating insights from both international relations and comparative politics it presents an updated theoretical framework on how external and do- mestic factors interact in shaping state-building. The rest of the article proceeds in five main sections. The next section discusses the state of previous research on the relationship between external military threats and state-building, and the politics of auto- cratic regimes. I then turn to my own theoretical argument and formulate two main hypotheses, followed by a description of methods and data. In section four, the main findings are presented. Lastly, I conclude with a discussion about the importance of the findings, possible limitations of the study, and present some avenues for future research. Previous research on external threats and state-building The idea that external threats lead to state-building is far from new. Seminal work by Tilly et al. (1975), as well as Downing (1992), Ertman (1997) and many others, emphasize how the threat of war helped to create strong, centralized states on the European subcontinent. The core argument is that the hostile geopolitical situation in Europe in the early modern period forced states to extract more and more revenue from their societies, since ever greater resources were needed to pay for better and more well-equipped armies. This, 3 in turn, created an incentive to control the state’s borders, build up a permanent bureaucracy and collect information about its population and economy. In order to not fall behind, other neighboring states in Europe were forced to do the same. Over time, thus, the pressure from adjacent states put in motion processes that lead to increased state power over their respective populations and territories. That said, the difficulties and obstacles to increasing state capacity should not be understated. Although most rulers would like to preside over a high-capacity state, they needed a strong impetus to build up such organizations. This is because the advent of more powerful states was often fiercely opposed by different groups in society (Migdal, 1988; Scott, 2009). The history of Europe’s early modern period is full of examples of societal resistance against growing state power. One main conclusion of this canonical literature is thus that there were clear costs associated with state-building. If rulers tried to tax their populations more heavily, rule over distant provinces directly, or step up the monitoring of their populations, people were likely to push back. Hence, the implicit argument in this line of state-building scholarship is that without the hostile interna- tional environment that characterized Europe, the costs associated with state-building would often out- weigh the potential benefits. Without external threats, much less state-building would have been attempted. While there is widespread agreement that the bellicist explanations for state-building in early modern Eu- rope are at least partly correct, there is far more disagreement among researchers as to the extent these mechanisms are present in the contemporary era (Taylor & Botea, 2008). Many scholars point out that the international order established after World War II differs in important aspects from that which came before. In the postwar period new norms about territorial integrity made violent border changes exceedingly rare (Zacher, 2001), and states could be afforded legal sovereignty even though they lacked effective control over their territories or populations (Jackson & Rosberg, 1982). This new international order was not only imposed on the rest of the world by the major powers, but was also supported by post-colonial rulers themselves1. Other researchers point out that governments in the contemporary era more easily can borrow money on international capital markets, thus partly limiting the need to raise taxes to fund increased military spending (Ahmed, 2012; Brzoska, 1983; Centeno, 2002; Queralt, 2019; Smyth & Kumar Narayan, 2009). The overall message from this body of literature is that in the world order established after 1945, it is unlikely that inter-state military conflicts – or the threat thereof – have a positive effect on state-building. This conclusion, though, is highly disputed. A number of researchers argue that the bellicist account of state-building is still applicable, and see the relative absence of external threats after 1945 as the main ex- planation for the existence of weak states in much of the global South (Desch, 1996; Migdal, 1988). More- over, in recent years a number of quantitative studies has demonstrated the continued relevance of external military threats as an impetus for state-building (Gibler & Miller, 2014; Han & Thies, 2019; Thies, 2007; Thies, 2004). In short, although violent border changes and existential wars between nation-states have become less common, there are strong reasons to believe that pressures and itimidation from other states still play a role in pushing rulers towards creating stronger states. What is missing in much of the quantitative literature focusing on state-building in the postwar era, how- ever, is theories about how international threats interact with domestic factors in producing state-building2. Although the studies vary in both temporal and geographical scope, they typically assume that the effect of such threats on state-building is uniform across cases. Thus, there is little theorization about if and how different regimes react differently to international threats. In this study, I aim to remedy this shortcoming by studying how fear of internal and external threats interact in shaping state-building in an important 1 For example, one of the founding principles of the Organization of African Unity – formed in 1963 by 32 newly independent African states – was that all territorial boundaries inherited from colonialism should be upheld (Herbst, 2000) 2 A few small-N studies have developed more nuanced argument as to when – or under which conditions – external threats can lead to state-building (Barnett, 1992; Doner et al., 2005). These studies, however, typically focus on a single country or a smaller region, without trying to make more general claims. 4 subset of cases – in autocracies. While scholars in recent years has argued about wether democracy is con- ducive to state-buidling or not – as in the so called “sequencing debate” (Carothers, 2007; Fukuyama, 2007; Mansfield & Snyder, 2007) – less attention has been placed on when and how state-capacity develops in non-democracies. Moreover, after a long period of democratization, autocracy around the world seem to be on the rise (Alizada et al., 2021). In the next section, I lay out my theoretical argument explaining how power-sharing institutions shape the extent to which autocracies respond to foreign threats with state- building efforts. Theory As the literature review in the previous section made clear, there is no agreement amongst researcher about how applicable “bellicist” accounts of state-building is in the contemporary era. This study, however, con- tends that there are strong reasons to believe that military threats from neighboring countries still play a role in pushing regimes towards creating stronger states. To begin with, governments – also after 1945 – closely follow what their neighbors do in military terms (Collier & Hoeffler, 2002). At various times in the last half century, countries in different regions of the world have engaged in what could be described as arms races – that is, ramping up their military capabilities in response to other nearby states doing the same (Heiduk, 2017; Kollias & Paleologou, 2002; Solmirano & Perlo-Freeman, 2009; Yesilyurt & Elhorst, 2017). Consider neighboring Iran and Iraq, who for decades tracked each other’s military capabilities – a develop- ment which has continued also after the devastating war of the 1980s. The same pattern is discernable around the world, as illustrated by adjacent Morocco and Algeria, Colombia and Venezuela, Malaysia and Indonesia (see figure 1), and in many other instances. FIGURE 1. MILITARY SPENDING FOR SELECTED COUNTRIES. 5 Such patterns are classic examples of a “security dilemma,” a cornerstones of realist theories of international relations (Jervis, 1978). In other words, there are strong reasons to believe that the activities of neighboring countries affect how many resources dictators pour into their own militaries. State-capacity, however, is about more than having a well-equipped army. A capable state can extract resources from society in an efficient manner, employs a capable bureaucracy that is present throughout the territory, and which collects information about its population and economy in a systematic and regular way. Such a state exhibits a high degree of “infrastructural power,” understood as “the institutional capacity of a central state ... to penetrate its territories and logistically implement decisions” (Mann, 1984; Soifer, 2008). One of the mechanisms highlighted in the canonical literature on state-building in Europe – demand for revenue – is still relevant today (Thies, 2007; Thies, 2004; Tilly, 1992). Autocrats who want to spend more on their military need money. In the short run, it is of course possible to reduce spending on other govern- mental projects in order to fund an increased defense budget, but this is likely unsustainable in the long run. Instead, increased taxation is needed to fund military equipment and advanced weapon systems3. Ef- ficient taxation in turn necessitates reliable statistical information about the economic actors in the country and hence a relatively competent bureaucracy. Rising demand for revenue is not the only mechanism linking external military threats to increased state capacity, however. A more well-funded and competent bureau- cracy enables the government to enact policies that are not strictly military in nature, but may nevertheless help to defend the country in the long run. South Korea is an illustrative example. There, the severe threat represented by its northern neighbor pushed the government to create a more cohesive “Weberian” bu- reaucracy – which could be used to not only increase taxes, but also to enact infrastructure projects and promote industrialization (Doner et al., 2005; Fields, 1997). Lastly, rearmament or saber-rattling from neighboring states provides an opportunity for autocratic regimes to paint themselves as the protector of the country and its people (von Soest & Grauvogel, 2017). Studies indicate that external military threats make citizens more willing to consent to resource extraction (Gibler & Miller, 2014), and that domestic opposition against the incumbent decreases during times of foreign intimidation (Gibler, 2010). Even relatively mild forms of threat made by foreign states increase the level of “pro-government mobilization” (Hellmeier, 2020) in autocracies. Thus, external threats against autocra- cies not only push regimes towards state-building, they also increase the likelihood that populations will accept the rise of stronger states without widespread revolts or protests. The politics of autocratic rule To govern over millions of people, any ruler needs the support of other influential figures. No matter how “all-powerful” an individual dictator may be, he still needs to rely on a wider elite group to govern. Much research on autocracies thus focuses on the relationship between the dictator and the elite group supporting him. Together, the regime elite has enough instruments of power to uphold the ruler’s position, and if they act collectively, they may also be able to remove the dictator from power. While the dictator tries to acquire more power at the expense of his allies, the elite group supporting the dictator constantly attempts to monitor the ruler so that he does not succeed in his attempt. This interplay between the dictator and his regime allies has been dubbed the “problem of authoritarian power-sharing” (Svolik, 2012) or as the ques- tion of “elite consolidation” (Geddes et al., 2018). This built-in tension between dictator and elite means that many autocrats are wary of threats from within. Looking at all dictators between 1946 and 2008, Svolik (2012) documents that more than two-thirds of all non-constitutional exits from power happened because regime insiders acted to remove the ruler. This 3 I come back to the question about borrowing in the next section. 6 pattern reveals a crucial difference between democracies and dictatorships. While conflicts between the executive and members of her cabinet or political party are commonplace also in democracies, the stakes associated with such conflicts are of a different order of magnitude in dictatorships. While a democratic leader who fails in her aspirations may be forced from office, a dictator who loses out in a political power- play is often faced with exile, imprisonment or death (Escribà-Folch, 2013). The importance of threats from within in autocracies has led researchers to focus on features that can mitigate such threats. A growing literature on autocratic politics argues that institutions such as legislatures, parties and advisory councils tend to stabilize dictatorships and prolong rulers’ tenure (Boix & Svolik, 2013; Gandhi & Przeworski, 2007; Geddes, 1999). One key reason is that such institutions alleviate the commit- ment and monitoring problems that tend to plague authoritarian regimes. The existence of routine negoti- ations between ruler and elites means that regime insiders can more easily detect if the dictator tries to acquire more power for himself. Such institutions thus serve as a kind of assurance for the individual dic- tator and members of the elite alike: Knowing that his allies will likely detect him if he tries to concentrate power to his own person, the dictator will refrain from doing so. For the same reason, members of the regime elite will be less suspicious of the dictator, meaning that regime infighting will be less likely, and less severe. From the perspective of the elite, actual influence over policy is thus always preferable to simply receiving club goods from the ruler. Even if political support in exchange for material benefits may seem like a good deal today, the elite cannot be certain that the dictator won’t change his mind in the future. Higher levels of institutionalized power-sharing – where regime elites get actual influence over important areas – mitigate such fears. Power-sharing and autocratic state-building I expect autocracies with a higher degree of institutionalized power-sharing to respond to rearmament in their regional neighborhood with more ambitious state-building efforts, compared to autocracies with little or no power-sharing. The reason is straight-forward: If state-building – that is, attempts to create an infra- structurally more powerful state – is seen as empowering the dictator personally, his regime allies will likely oppose such a move, and may even try to remove him from power. It is paramount to understand that the members of the regime elite might very well oppose the rise of a stronger state even if they themselves are not subject to the state’s power. The reason is that even if members of the regime elite are exempt from state authority today, they cannot be certain that the state’s newly acquired powers will not be used against them tomorrow (Andersson, 2019). This is why power-sharing institutions matter (Boix & Svolik, 2013; Gandhi & Przeworski, 2007) for state- building. If a legislature, politburo or military counsel has the authority to make key decisions, members of this institution will be less fearful of an infrastructurally more powerful state. If members of the regime elite rule the state together with the dictator, the risk that the state’s powers are used against them is smaller. Anticipating that opposition from regime elites will be weaker, the dictator will thus dare to invest in state capacity. On the other hand, when institutionalized power-sharing is absent, regime insiders will likely try to block the rulers’ attempts to build up a stronger state apparatus. Thus, suspicion and distrust within the regime will be more widespread. In such circumstances, despite foreign threats, investment in state capacity is likely to be low or nonexistent. It should be underscored that even in those instances where regime elites cannot stop attempts to create more capable state institutions, their opposition to state-building may never- theless make such attempts futile (Boone, 2003; Kurtz, 2013; Soifer, 2015). Again, such opposition is less likely if elite groups are included in the decision-making processes about how the state is run. In the face of external threats, power-sharing also affect state-building through two other channels: By impacting the character of the bureaucracy, and by affecting how military costs are financed. A permanent 7 bureaucracy is an integral part of a high-capacity state. But as a dictator becomes more reliant on his ad- ministration, he also faces the possibility that people within the bureaucracy use their expertise and position to depose him. If they ally themselves with a challenger to the ruler, their position and knowledge may be the factor that tips the scales in the rivals’ favor. Many dictators thus fear competent bureaucrats. The more capable and knowledgeable they are, the bigger the potential threat. For this reason, autocrats tend to prefer incompetent – but loyal – subordinates (Egorov & Sonin, 2011). While rewarding loyalty over competence can make the dictator more secure, at least in the short run, it also results in a badly functioning bureaucracy. In situations in which a dictator single-handedly can remove an advisor who displeases him, bureaucrats will also be fearful of telling the truth to him (Frantz & Ezrow, 2011). Thus, many autocrats surround themselves with “yes men”: persons who only tell the dictator what they think he wants to hear. As this culture of fear and bad information (Papaioannou & Van Zanden, 2015) is translated down through the bureaucratic system, the consequence is an inefficient and non-cohesive state apparatus. However, power- sharing institutions may help to reduce the fear and mistrust that plagues many autocracies. Dictators in such systems will be less fearful of treason, and the “competence-loyalty trade-off” less severe. If power- sharing is substantial, the individual dictator will likely not be able to single-handedly remove a high-ranking administrator even if he displeases him. While a bureaucrat who falls out of favor with the ruler in a “per- sonalist” (Geddes, 1999) regime – where the ruler makes key decisions without consultation with others – often risk dismissal, exile or even death, this is less often the case under constrained dictatorships. Where arbitrary dismissals are less prevalent, bureaucrats tend to be more competent (Nistotskaya & Cingolani, 2015). Lastly, power-sharing also affects how costs associated with military buildups are funded. If military spend- ing is paid for by taking on foreign debt rather than by revenue-raising, the link between external threats and state-building weakens. Although results are mixed, studies indicate that there is a positive association between increased military expenditure and foreign debt, at least in parts of the global South (Brzoska, 1983; Smyth & Kumar Narayan, 2009). Borrowed money, however, usually needs to be paid back, meaning that they can be seen as “deferred taxes”. Thus, there are strong reasons to expect short-sighted rulers to prefer borrowing, and those with longer time-horizons to prefer tax increases (Queralt 2019). A regime uncertain of whether it will remain in power in ten years’ time will likely not bother to build up fiscal capacity if sufficient international credit is available instead. By contrast, autocracies with more constrained dictators are more stable, and are thus more likely to have longer time-horizons (Geddes, 1994). Moreover, taxation typically involves some political concessions. For this reason, low degrees of initial power-sharing make it more likely that the ruler opts for borrowing instead of taxation (Queralt, 2019, p. 714). . Summary and hypotheses To summarize, I argue that autocracies respond to military buildup in neighboring countries by increased state-building efforts, but that this association is moderated by the degree of institutionalized powers-shar- ing within the regime. Where the ruler is constrained by the wider regime elite, fear and mistrust is less prevalent, time-horizons longer, and infighting less severe. Such regimes are more likely to answer potential foreign military threats with investment in state capacity. Two testable hypotheses can be derived from this theory: 8 H1: In authoritarian regimes, increased external military threats are associated with increased state capacity. H2: The association between external military threats and state capacity is stronger in authoritarian regimes with higher levels of institutionalized power-sharing. Data and method The research design takes advantage of time-series cross-sectional data covering over a hundred countries in the postwar era4. I run fixed effect linear regression models, which by default hold all group-specific, time-invariant factors constant. This means that in the analysis all country-level factors that do not change over time are automatically controlled for. This includes (but goes beyond) geographical features, colonial origins, ethnic fractionalization, culture, and so on. The fact that fixed effect regressions wipe out all group- specific unobserved heterogeneity allows the researcher to identify causal effects under weaker assumptions than standard regression models (Brüderl & Ludwig, 2014, p. 327). In addition to controlling for time- invariant group-specific features, all models also include year-fixed effects, allowing to control for aggregate time-series trends and shocks. To alleviate concerns of autocorrelation and heteroskedasticity, all models use robust standard errors, clustered by country. Dependent Variable Although research on state capacity has exploded in recent years (Cingolani, 2013; Hendrix, 2010), good measurements on state capacity are few and far between. Some measures are theoretically sound, but suffer from limited geographical (Brambor et al., 2019) or temporal (Lee & Zhang, 2017) scope. Other measures, such as the one developed by D’Arcy and Nistotskaya (2017), have both great temporal and spatial cover- age, but focus only on a specific component of state capacity (cadastral records). It is not unusual for researchers who claim to be studying state capacity to in fact use measures that rather tap into concepts such as “quality of government” (Charron & Lapuente, 2011; Rothstein & Teorell, 2008), GDP p.c. (Fearon & Laitin, 2003), level of corruption (Bäck & Hadenius, 2008) or the amount of private credit (Besley & Persson, 2009). Rather than using any single indicator as a proxy for state capacity, I take advantage of the latent Capacity- index created by Hanson & Sigman (2013), based on a widely used definition of state capacity as “the ability of state institutions to effectively implement official goals” (Hanson & Sigman, 2013, p. 2). Hanson and Sigman focus on three core dimensions of state power – coercive, extractive and administrative capacity – arguing that these dimensions correspond to what Skocpol (1985) calls the “general underpinnings of state capacities”: stable administrative-military control of a given territory, skilled officials, and plentiful financial resources (Lindvall & Teorell, 2016). While it is theoretically possible to distinguish between these three dimensions of state capacity, they are very hard to empirically disentangle from one another. There are good reasons to believe that all three dimensions are highly interrelated and mutually supporting. In order to extract revenue, the state must have military-administrative control over its territory and population. Coercive capacity requires financial re- sources. A competent bureaucracy makes taxation easier, and so on (Hanson & Sigman, 2013, p. 4-5). Using Bayesian latent variable analysis that employs 24 different indicators5, Hanson and Sigman creates an index 4 The sample is described in more detail further below. 5 A list of the 24 indicators is found in the appendix. 9 covering up to 162 countries for the time-period 1960-2009. Having been cited well over two hundred times, the index has been used in empirical research by a large number of scholars, including Bizzarro et al. (2018), Wang & Xu (2018), Guillén & Capron (2016), Houle (2017) and many others. The resulting Capacity-index is standardized to have a mean of 0 and a standard deviation of 1. In the full dataset, the highest value for any country-year is 2.86 (Singapore in 2008), and the lowest value is -3.51 (Somalia in 2007). While these positions at the respective extreme ends of the scale are relatively uncontro- versial, figure 2 shows the development of state capacity for four other countries – Libya, Malaysia, China and Haiti – which for long periods have been ruled by various autocratic regimes. The four countries display vastly different trajectories, indicating that the index has not only great between-country variation, but also captures variation within countries over time. State capacity in Libya, for example, shows a V-formed pat- tern. During the 1960s and 70s, the North African country managed to increase state power markedly. After a coup in 1969 that deposed king Idris I, Muammar Gaddafi governed the country for more than forty years as an archetypical “sultanistic” ruler (Chehabi & Linz, 1998), with little or no de facto power- sharing. In the late 1970s Libya was involved in military conflicts with both Chad and Egypt. Starting around 1980, state capacity in the country began to erode. This downturn trend roughly coincides with Gaddafi’s implementation of the Jamahiriya system – literally meaning “a state of the masses” – which from 1977 onwards entailed a massive overhaul of the bureaucracy. Meritocratic recruitment to top positions in the bureaucracy stopped, and the public administration became almost obsolete (Sawani, 2018). FIGURE 2. STATE CAPACITY OVER TIME China tells a different story. In the late 1950s a crack began to appear between China and the USSR. In the 1960s relations were openly hostile, culminating in violent border clashes in 1969. State capacity, however, remained relatively low during the 1960s. Following Mao’s “cultural revolution” the state’s capabilities even decreased somewhat during the 1970s and early 1980s. After Mao’s death, regime infighting became less 10 severe, and the communist party returned to a system of collective rule. Around 1980, under Deng Xiao- ping, the party began to invest massively in education and research, and modernized key governmental agencies (Evans, 1997). A few years later, state-capacity began to rise. The above descriptions illustrate well that the index is not merely a reflection on one aspect of state power. For example, while Libya had a fairly big military in 2009 – corresponding to about 1.3 % of the population – its bureaucracy was functioning badly6. Thus, while Libya’s coercive powers were substantial, its public administration was inefficient and its fiscal capacity severely underdeveloped. By contrast, in 2009 Malaysia was endowed with a competent bureaucracy, and received considerable amounts of revenue from income- taxes; indicating a sophisticated extractive apparatus. As Figure 1 makes apparent, Haiti scores badly on all dimensions of state capacity. Independent Variable Once they have broken out, wars or invasions are fairly easy to measure. It is much harder, however, to gauge the threat of a military conflict. This is the case because threats are latent; the possibility of something that could happened. Modern research on the effect of external threats on state-building largely deals with this problem in two different ways. Some researchers (Han & Thies, 2019) take advantage of expert-based evaluations on inter-state rivalries (Thompson, 2002). Alternatively, researchers have used data on actual military conflicts (Gibler, 2010; Gibler & Miller, 2014), ranging from full-fledged wars to less severe milita- rized disputes. Although these measurements all have their merits, I propose a new way to gauge the potential military threat faced by each state in the international system: as the sum of military spending in adjacent countries (log transformed). Adjacent countries stand for all countries that share a land-border with a given state at any particular year. Data on military spending is taken from the Correlates of War project (Singer, 1988). This operationalization constitutes an improvement on existing measures in several ways. Firstly, it taps into the concept of “military threat” better than the military conflict measure, and unlike indicators of inter- state rivalries, it relies on data that is more objective or “hard.” Secondly, both measurements based on expert-evaluations and on actual military incidents result in highly skewed variables with little variation. Typically, in most existing studies military threat is operationalized by dichotomous variables: For a given country- or dyad-year, one either observe, or do not observe, a strategic rivalry or military incident. Such an operationalization of military threat leads to the impression that for most of the time a large majority of all countries in the postwar era faced no (i.e., “zero”) external threats whatsoever. The third advantage of this measure is that by taking the sum of military spending for all adjacent countries, geographical proximity is accounted for. This matter because – everything else being equal – the closer a country is located to another country, the greater the potential military threat. This is the case because it is extremely costly to move troops, ships and aircrafts across great geographical distances (Boulding, 1962). Furthermore, a shared land-border means that military units can move into the other country without vio- lating another state’s territory. This further underscore the importance of adjacent countries, as opposed to countries that are merely located in the same region. The concept of a “buffer state” (Partem, 1983) captures this idea well. One possible objection to this operationalization is that it doesn’t pay enough attention to military alliances. If a neighboring country is an ally, it’s military spending should not be considered a threat. For this reason, 6 For example, the Government Effectiveness indicator provided by the World Bank – on which the scale goes from -2.5 to 2.5 – codes Libya in 2009 as “-1.” 11 I also create an alternative measure, which takes the two most important military alliances of the post-war era – NATO and the Warsaw-pact – into account. If two adjacent countries share membership in either of these alliances, their military outlay will be excluded when calculating the military expenditures of all neigh- boring countries7. FIGURE 3. MILITARY SPENDING IN ADJACENT COUNTRIES OVER TIME Figure 3 shows how military spending amongst neighbors has changed over time for four selected countries. Although there is considerable variation both within and between countries, there is a clear overall trend of increased military spending. However, we can also discern periods where military spending in adjacent countries drop considerably – sometimes very sharply. For example, in the early 1990s military spending amongst Chinas neighbors fell off a cliff. Obviously, this is related to the breakup of the Soviet Union, which had a big negative impact on Russia’s military spending. This highlights the importance of controlling for aggregate time-series trends and shocks. Moderating Variable How to operationalize power-sharing in autocracies is far from straight-forward. Much earlier work uses dummies of authoritarian parties or parliaments (see for example Gandhi & Przeworski 2006; Svolik & 7 As a further robustness check, I also interact the main independent variable – the sum of military spending in adjacent countries, log transformed – with a dummy variable, which takes the value 1 if a country is a member of either NATO or the Warsaw-pact in that particular year. 12 Boix 2013) to describe situations where the dictator is constrained by a broader coalition. While this results in transparent coding-rules, it is arguably also very blunt. As Meng (2021) points out, far from all parties in autocracies have the capacity to actually constrain the ruler. Strøm et al. (2017) presents another way to measure power-sharing in both dictatorships and democracies. Although more nuanced than simple dum- mies, their measures suffer from some of the same limitations in that it primarily captures de jure, rather than de facto power-sharing institutions (Strøm et al., 2017, p. 171). This can be especially problematic in autocracies, where the difference between official rules and actual reality if often large8. For this reason, I rely on the “executive constraints” variable provided by Polity (Marshall et al., 2013), built around experts’ assessment of the actual level of power-sharing. The values of the variable – which captures “the extent of institutionalized constraints on the decision-making powers of chief executives” (2013, p. 24) – range from one to seven, with higher values indicating a higher degree of institutionalized power- sharing. Autocracies where the executive constraints variable is coded as “1” include regimes such as Saudi Arabia in the postwar era, Uzbekistan after the breakup of the USSR or Libya under Muammar Gaddafi. Non-democracies where the same variable is coded as “7” include Malaysia for parts of the 1960s, South Africa during apartheid or Egypt up to the 1952 revolution. Regimes in the middle of the spectrum include China from 1976 onwards (3), Iran in the 1990s (3-4), and Mexico during the reign of the PRI up to the late 1990s (3-4). It is worth highlighting that constraints on the executive could also be measured using new data from the V-Dem Institute. It’s “judicial constraints on the executive index” and the “legislative constraints on the executive index” both have great geographical and temporal coverage (Coppedge et al., 2020). However, the V-Dem indicators suffer from the fact that they only cover specific forms of constraints on the ruler. By contrast, a strength with Polity’s measure is that it is not restricted to a particular type of power-sharing institution. Instead, it aims to measure executive constraints regardless of whether power is shared between the ruler and a parliament, a ruling party, a military junta, or something else. Controls Given the two-way fixed effects setup, I am conservative in my selection of further control variables. The only additional controls included in the models are those for which there is a strong theoretical reason to believe that they causally affect both the independent and dependent variable (i.e., “confounders”). Apart from country- and year-fixed effects, I control for GDP per capita, recent history of inter-state armed conflict, and oil production. Controlling for GDP per capita is important, since a country with more eco- nomic resources poses a bigger threat to its neighbors, thus potentially influencing military spending in adjacent countries. At the same time, richer countries can devote more resources to the creation of a capable state apparatus. The history of inter-state war is captured by a dummy variable that takes the value one if any adjacent country has been involved in an inter-state armed conflict (Pettersson & Eck, 2018) in the last five years. Having war-prone neighbors is expected to lead to increased state-building efforts, but it is also likely to have a causal impact on military spending for all countries in the region. Lastly, oil-rich nations represent a bigger threat to their neighbors than those without oil reserves, thus affecting military spending in adjacent countries. Simultaneously, most researchers argue that abundant oil reserves have a negative effect on state 8 Consider the variable “constraining power-sharing” constructed by Strøm et al. (2017), which attempts to measure the extent to which institutions can constrain politicians from actions deemed “predatory” against vulnerable individuals or social groups. Ac- cording to the measure, politicians in Sierra Leone in the 1980s where more constrained than those in Sweden, the United States or the Netherlands. At the same time, electoral violence in Sierra Leone was common, and the country’s dictator Siaka Stevens was fond of saying that “force is the only language the ordinary man understands” (Kandeh 1999, p. 359). 13 capacity (Brautigam et al., 2008). The variables capturing oil wealth (Ross & Mahdavi, 2015) and GDP per capita (Gleditsch, 2002) are both log transformed. TABLE 1. SUMMARY STATISTICS Variable Obs Mean Std.Dev. Min Max State capacity 3814 -.363 .806 -3.512 2.22 Neighbours’ military expenditure (log) 3814 14.558 2.459 8.12 19.637 Executive constraints 3550 2.284 1.382 1 7 Neighbor war history 3814 .33 .47 0 1 GDP p.c. (log) 3814 7.778 1.07 4.889 13.357 Oil wealth (log) 3813 9.7 10.53 0 26.357 Sample To enter the sample, a country-year must be coded as a non-democracy by Boix et al. (2012). Due to data limitations for the Capacity-index; the maximum time-period covered for any given country is 1960 to 2009. Given the operationalization of military threat as the sum of military spending in all adjacent countries, island states without any shared land borders are excluded from the sample. These rules produce a sample of 3 814 country-years9. A full list of the countries and years included in the sample (table A3) is found in the appendix. Results In the first step of the analysis, I investigate hypothesis H1. Results are displayed in table 2 below. All models include country- and year-fixed effects, with standard errors clustered by country. Model 1 depicts the bivariate association between neighboring countries military spending and state capacity, excluding any controls bar country- and year-fixed effects. As hypothesized, the coefficient for military spending is statis- tically significant and signed positively. The magnitude of the “effect,” however, is somewhat difficult to interpret. Technically, a one percent increase in military spending is associated with a 0.000782 increase in state capacity. A more meaningful interpretation suggests that if military spending in the surrounding region is increased by one standard deviation – roughly what Libya experienced between the mid-1950s and early 1970s – the Capacity-index would increase by roughly 0.19. As the mean for the Capacity-index in this sample is -0.36, with a standard deviation of 0.81, this is a relatively small effect. This finding is in line with the literature that argues that state-building today is a slow and gradual process (Ayoob, 1995; Thies, 2007). Model 2 introduces some time-varying control variables. The coefficient for neighboring countries’ military spending remains positive, and highly significant. Controlling for GDP per capita, oil-production and re- gional war-history somewhat strengthens the relationship between the independent and dependent variable. Model 3 and model 4 replicate model 1 and 2, but uses the alternative independent variable described in the previous section: If two adjacent countries share membership in either NATO or the Warsaw-pact, their military outlay are excluded when calculating the military expenditures of all neighboring countries. 9 Because of missing data on some covariates, the actual number of observations drop somewhat when including more predic- tors. 14 As can be seen, the use of this alternative independent variable does not change the results in any substantial way. All in all, the results presented in table 3 provide clear support for hypothesis H1. Military buildups in neighboring countries are associated with higher levels of state-capacity. TABLE 2. THE EFFECT OF NEIGHBORS’ MILITARY EXPENDITURE ON OWN STATE CAPACITY Model 1 Model 2 Model 3 Model 4 Neighbours’ military expenditure (log.) 0.0782** 0.100*** (0.0293) (0.0284) Neighbours’ military expenditure (log.) alternative 0.0774** 0.0973*** (0.0284) (0.0272) Constant -1.859*** -4.327*** -1.833*** -4.275*** (0.351) (0.516) (0.335) (0.503) Country-fixed effects Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes Controls No Yes No Yes N 3786 3736 3786 3736 Countries 117 117 117 117 Time period 1960-2009 1960-2009 1960-2009 1960-2009 Within R2 0.279 0.334 0.279 0.333 Between R2 0.212 0.617 0.0714 0.477 Overall R2 0.226 0.572 0.135 0.478 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy). The dependent variable in all models is state capacity. * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. Hypothesis H2 – that the association between external military threats and state capacity is stronger in more constrained autocracies – is investigated in the next group of models, reported in table 3. In model 5 the effect of neighbors’ military spending on state capacity is allowed to vary depending on the degree of power- sharing, but leaves out additional controls. The interaction term is significant at the 99% level, and positively signed. Model 6 includes the same interaction-term as model 5, and also controls for GDP p.c., oil wealth and war-history of neighboring countries. Now, the interaction-term is significant at the 99.9% level, and the size of the coefficient is somewhat bigger than in model 5. 15 TABLE 3. THE INTERACTION BETWEEN NEIGHBORS’ MILITARY SPENDING AND EXECUTIVE CON- STRAINTS ON STATE CAPACITY Model 5 Model 6 Model 7 Model 8 Neighbours’ military expenditure (log.) 0.0313 0.0572* (0.0262) (0.0226) Executive constraints -0.213** -0.233*** -0.194** -0.207*** (0.0708) (0.0577) (0.0697) (0.0563) Neighbours’ military expenditure (log.) ## 0.0146** 0.0160*** Executive constraints (0.00486) (0.00418) Neighbours’ military expenditure (log.) alternative 0.0364 0.0607** (0.0253) (0.0219) Neighbours’ military expenditure (log.) alternative ## 0.0133** 0.0142*** Executive constraints (0.00478) (0.00405) Constant -1.218*** -3.445*** -1.268*** -3.480*** (0.328) (0.473) (0.313) (0.459) Country-fixed effects Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes Controls No Yes No Yes N 3547 3498 3547 3498 Countries 117 117 117 117 Time period 1960-2009 1960-2009 1960-2009 1960-2009 Within R2 0.373 0.425 0.371 0.423 Between R2 0.172 0.584 0.0575 0.419 Overall R2 0.234 0.594 0.143 0.469 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy).The dependent variable in all models is state capacity. * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. Just as when investigating hypothesis H1 above I rerun the regressions using the alternative measure of neighboring countries’ military spending. The results are displayed in model 7 and 8. Again, the results stay largely the same. The size of coefficient for the interaction-term becomes somewhat smaller when using the alternative independent variable, but is still significant at the 99,9% level, and signed in the theoretically expected direction. 16 FIGURE 4. THE MODERATING EFFECT OF INSTITUTIONALIZED POWER-SHARING Estimated from model 6 in table 3, with 95 % confidence intervals. To better grasp the size of the interaction effect, the relationship between neighbors’ military spending and state capacity is plotted in Figure 4, at the minimum and maximum values of executive constraints. As can be seen in the figure, the slope of the two lines differs rather dramatically. For unconstrained dictatorships, a one standard deviation increase in adjacent countries’ military spending is associated with an increase in the Capacity-index of roughly 0.18. However, in autocracies with the highest degree of power-sharing, the same upturn in neighboring countries’ military expenditure is associated with an increase in the Capacity- index of about 0.42. To get a sense of what an increase in the Capacity-index of 0.42 means, we can examine the level of state capacity in two African countries in 2009. The Capacity-index for Nigeria in West Africa was about -0.97, which can be contrasted with East-African Rwanda’s Capacity-index in the same year (- 0.53). Thus, a 0.42 increase in the Capacity-index is about as big as the difference between Nigeria and Rwanda in 2009. While both countries are relatively underdeveloped by global standards, the Rwandan state is often considered somewhat of an “outlier” in terms of state capacity. Many observers have noted that despite being fiscally poor, Rwanda possess a high degree of infrastructural power (Soifer, 2008; Straus, 2006). As a further comparison, in 2009 the World Bank ranked Rwanda 68th out of 191 countries in terms of control of corruption – ahead of countries such as Croatia and Greece – and 91st from the top in terms of government effectiveness. Nigeria, by contrast, was ranked 157th and 173rd for the same two measures. To sum up, the results presented in table 4 provide clear support for hypothesis H2. While an upturn in surrounding states’ military spending is associated with an increase in state capacity, this effect is especially pronounced in autocracies with a high degree of power-sharing. 17 Robustness In addition to the models presented in table 2 and table 3, I conduct several robustness checks. To make sure that the results are not driven by a particular categorization of non-democracies, I replicate the study using alternative ways to select the sample. Running the main models using Geddes et al. (2014) classifica- tion of autocratic regimes rather than relying on the coding by Boix et al. (2012) does not alter the main findings substantively. The same is true when using different cut-off points of the polyarchy-index (Coppedge et al., 2020) to discriminate between observations, or when I exclude all autocratic country- years from Europe from the analysis. Although the size of the main coefficients varies somewhat, in all models the relevant coefficients are statistically significant, and signed in the theoretically expected direction (table A4-A8). I also interact the main independent variable – the sum of military spending in all neighbor- ing countries – with a dummy which takes the value 1 if a country-year is a member of either NATO or the Warsaw-pact (table A9). Again, results change very little. Moreover, I control for two of the 24 indicators that make up the Capacity-index – military expenditures p.c. (log), and military personnel p.c. – given that military spending in neighboring countries could affect these directly. In some models, the size of the rele- vant coefficient decreases somewhat, but they remain statistically significant and signed in the theoretically expected direction throughout (table A10-A11). To mitigate concerns that the executive constraints variable is in fact only measuring the level of democracy (Gleditsch & Ward, 1997), I control for two components of democracy that are not tapping into this concept: suffrage and free and fair elections (Coppedge et al., 2020), with results staying substantively the same (table A12). I also employ two alternative measures of power-sharing: the judicial constraints and the legislative con- straints on the executive index, both from V-Dem (Coppedge et al., 2020). In both cases, the interaction- term is signed in the theoretically expected direction, although the interaction-term including legislative constraints on the executive is not statistically significant (table A13). Given the fact that parliamentary checks on the ruler are relatively rare in modern dictatorships, this should not come as a great surprise10. I also check for outliers, since the results could be driven by a few unusual cases with the combination of very high levels of external threat and strong executive constraints. Dropping such outliers from the sample completely, as well as including an outlier dummy in the model, do little to change the overall picture (table A14-A16). The results of the robustness checks, as well as additional descriptive information, are available in the appendix. Conclusion In this article I improve on the orthodox interpretation of the “bellicist” theory for state-building by sug- gesting that external threats and domestic institutions interact in motivating rulers to invest in state capacity. Drawing on research from international relations and autocratic politics, I argue that dictatorships with little or no institutionalized power-sharing are less likely to respond to foreign threats with successful state- building, and that the opposite is true for autocracies with highly constrained executives. I empirically test two hypotheses derived from this theory by creating a novel indicator of military threats and leveraging it against a measure of state capacity under demanding controls in a dataset covering nearly all autocratic country-years in the world between 1960 and 2009. The data provided strong support for both testable 10 As discussed previously, modern dictators can be constrained in many different ways, and a potential weakness with the indicators provided by V-Dem is that they only measure a particular type of power-sharing. 18 propositions: A neighbors’ military expenditure matter for one’s own state capacity, and the effect on state capacity is stronger in autocracies with higher institutionalized power-sharing. The article makes a number of contributions to the literature. Apart from the theoretical synthesis, it also introduces a novel measure of military threats that is both intuitive and exhibits variation both within and between countries. Secondly, by treating state capacity as the dependent variable it fills an important gap in the research on autocratic politics that has so far avoided studying the causes of state-building in autocracies systematically. Since it typically is through state organizations that autocratic regimes extract revenue, quell public discontent and implement economic policies (Slater & Fenner, 2011), understanding why only some autocrats successfully build up capable states is paramount. However, a number of aspects of this study warrant further development. To begin with, the empirical investigation is subject to the same problems as all studies relying on observational data. Although two-way fixed effect regressions represent a tough “test” for the theory, endogeneity remains a concern. Moreover, the origins of institutionalized power-sharing are not addressed in the article. The theoretical framework presented here could potentially benefit from studies treating power-sharing as the dependent variable (Boix & Svolik, 2013). In recent years, a number of developments have put in question the assertion that border wars are a thing of the past. Russia’s invasion of Ukraine, Azerbaijan’s take-over of Nagorno-Karabakh, and growing ten- sions over territorial disputes in the South China Sea, all point to a future in which national boundaries are once again contested and militarized. Notably, these developments occur while autocracy globally seems to be on the rise (Lopes da Silva, 2020; Lührmann et al., 2020). Studying how dictators respond to foreign military threats may be ever more important in the years to come. 19 References Ahmed, A. D. (2012). Debt Burden, Military Spending And Growth In Sub-Saharan Africa: A Dynamic Panel Data Analysis. Defence and Peace Economics, 23(5), 485-506. https://doi.org/10.1080/10242694.2011.627163 Alizada, N., Col, R., Gastaldi, L., Grahn, S., Hellmeier, S., Kolvani, P., . . . Lindberg, S. I. (2021). Autocratization Turns Viral (Democracy Report., Issue. https://www.v- dem.net/static/website/files/dr/dr_2021.pdf Andersson, P. (2019). Fiscal Capacity in Non-Democratic States (8). Department of Political Science, Lund University. Besley, T., & Persson, T. (2009). The Origins of State Capacity: Property Rights, Taxation, and Politics. American Economic Review, 99(4), 1218-1244. https://doi.org/10.1257/aer.99.4.1218 Bizzarro, F. G., John Knutsen, Carl Henrik Hicken, Allen Bernhard, Michael Skaaning, Svend-Erik Coppedge, Michael Lindberg, Staffan I. (2018). Party Strength and Economic Growth. World Politics, 70(2), 275-320. https://doi.org/10.1017/S0043887117000375 Boix, C., Miller, M., & Rosato, S. (2012). A Complete Data Set of Political Regimes, 1800–2007. Comparative Political Studies, 46(12), 1523-1554. https://doi.org/10.1177/0010414012463905 Boix, C., & Svolik, M. W. (2013). The Foundations of Limited Authoritarian Government: Institutions, Commitment, and Power-Sharing in Dictatorships. The Journal of Politics, 75(2), 300-316. https://doi.org/https://doi.org/10.1017/S0022381613000029 Boone, C. (2003). Political topographies of the African state : territorial authority and institutional choice. Cambridge University Press. Boulding, K. E. (1962). Conflict and defense; a general theory. Harper. Brambor, T., Goenaga, A., Lindvall, J., & Teorell, J. (2019). The Lay of the Land: Information Capacity and the Modern State. Comparative Political Studies, 53(2), 175-213. https://doi.org/10.1177/0010414019843432 Brautigam, D., Fjeldstad, O.-H., & Moore, M. (2008). Taxation and state-building in developing countries : capacity and consent. Cambridge University Press. Brzoska, M. (1983). Research Communication: The Military Related External Debt of Third World Countries. Journal of Peace Research, 20(3), 271-277. https://doi.org/10.1177/002234338302000308 Bäck, H., & Hadenius, A. (2008). Democracy and State Capacity: Exploring a J-Shaped Relationship. Governance, 21(1), 1-24. https://doi.org/https://doi.org/10.1111/j.1468-0491.2007.00383.x Carothers, T. (2007). How Democracies Emerge: The “Sequencing” Fallacy. Journal of Democracy, 18(1), 12- 27. https://doi.org/doi.org/10.1353/jod.2007.0002 Centeno, M. A. (2002). Blood and debt : war and the nation-state in Latin America. Pennsylvania State University Press. 20 Charron, N., & Lapuente, V. (2011). Which Dictators Produce Quality of Government? Studies in Comparative International Development, 46(4), 397-423. https://doi.org/10.1007/s12116-011-9093-0 Chehabi, H. E., & Linz, J. J. (1998). Sultanistic regimes. Johns Hopkins University Press. Cingolani, L. (2013). The State of State Capacity: a review of concepts, evidence and measures. MERIT Working Papers(2013-053), 1-56. (United Nations University - Maastricht Economic and Social Research Institute on Innovation and Technology (MERIT)) Collier, P., & Hoeffler, A. (2002). Military Expenditure: Threats, Aid, and Arms Races World Bank: Policy Research Working Papers, 2927, 1-23. https://doi.org/https://doi.org/10.1596/1813-9450-2927 Coppedge, M., Gerring, J., Knutsen, C. H., Lindberg, S. I., Teorell, J., Altman, D., . . . Daniel, Z. (2020). V- Dem [Country–Year/Country–Date] Dataset v10. https://doi.org/https://doi.org/10.23696/vdemds20 D'Arcy, M., & Nistotskaya, M. (2017). State First, Then Democracy: Using Cadastral Records to Explain Governmental Performance in Public Goods Provision. Governance, 30(2), 193-209. https://doi.org/https://doi.org/10.1111/gove.12206 Desch, M. C. (1996). War and Strong States, Peace and Weak States? International Organization, 50(2), 237- 268. Doner, R. F., Ritchie, B. K., & Slater, D. (2005). Systemic Vulnerability and the Origins of Developmental States: Northeast and Southeast Asia in Comparative Perspective. International Organization, 59(2), 327-361. Downing, B. M. (1992). The military revolution and political change : origins of democracy and autocracy in early modern Europe. Princeton University Press. Egorov, G., & Sonin, K. (2011). Dictators and Their Viziers: Endogenizing the Loyalty–Competence Trade-off. Journal of the European Economic Association, 9(5), 903-930. https://doi.org/10.1111/j.1542- 4774.2011.01033.x Escribà-Folch, A. (2013). Repression, political threats, and survival under autocracy. International Political Science Review, 34(5), 543-560. https://doi.org/10.1177/0192512113488259 Evans, R. (1997). Deng Xiaoping and the making of modern China (Rev. ed.). Penguin Books. Fearon, J. D., & Laitin, D. D. (2003). Ethnicity, Insurgency, and Civil War. The American Political Science Review, 97(1), 75-90. Fields, K. (1997). Strong States and Business Organization in Korea and Taiwan. In Business and State in Developing Countries. In B. Schneider & S. Maxfield (Eds.), Business and State in Developing Countries. Cornell University Press. Frantz, E., & Ezrow, N. M. (2011). The politics of dictatorship : institutions and outcomes in authoritarian regimes. Lynne Rienner Publishers. Fukuyama, F. (2007). Exchange: Liberalism Versus State-Building. Journal of Democracy, 18(3), 10-13. https://doi.org/doi:10.1353/jod.2007.0046 21 Gandhi, J., & Przeworski, A. (2007). Authoritarian Institutions and the Survival of Autocrats. Comparative Political Studies, 40(11), 1279-1301. https://doi.org/10.1177/0010414007305817 Geddes, B. (1994). Politician's dilemma : building state capacity in Latin America. University of California Press. Geddes, B. (1999). What Do We Know About Democratization After Twenty Years? Annual Review of Political Science, 2(1), 115-144. https://doi.org/10.1146/annurev.polisci.2.1.115 Geddes, B., Wright, J., & Frantz, E. (2014). Autocratic Breakdown and Regime Transitions: A New Data Set. Perspectives on Politics, 12(2), 313-331. https://doi.org/10.1017/S1537592714000851 Geddes, B., Wright, J., & Frantz, E. (2018). How dictatorships work : power, personalization, and collapse. Cambridge University Press. Gibler, D. M. (2010). Outside-In: The Effects of External Threat on State Centralization. Journal of Conflict Resolution, 54(4), 519-542. https://doi.org/10.1177/0022002710370135 Gibler, D. M., & Miller, S. V. (2014). External territorial threat, state capacity, and civil war. Journal of Peace Research, 51(5), 634-646. https://doi.org/10.1177/0022343314531003 Gleditsch, K. S. (2002). Expanded Trade and GDP Data. Journal of Conflict Resolution, 46(5), 712-724. https://doi.org/10.1177/0022002702046005006 Gleditsch, K. S., & Ward, M. D. (1997). Double Take:A Reexamination of Democracy and Autocracy in Modern Polities. Journal of Conflict Resolution, 41(3), 361-383. https://doi.org/10.1177/0022002797041003002 Guillén, M. F., & Capron, L. (2016). State Capacity, Minority Shareholder Protections, and Stock Market Development. Administrative Science Quarterly, 61(1), 125-160. https://doi.org/10.1177/0001839215601459 Han, E., & Thies, C. (2019). EXTERNAL THREATS, INTERNAL CHALLENGES, AND STATE BUILDING IN EAST ASIA. Journal of East Asian Studies, 19(3), 339-360. https://doi.org/10.1017/jea.2019.24 Hanson, J. K., & Sigman, R. (2013). Leviathan's Latent Dimensions: Measuring State Capacity for Comparative Political Research APSA 2011 Annual Meeting Paper, https://ssrn.com/abstract=1899933 Heiduk, F. (2017). An arms race in Southeast Asia? Changing arms dynamics, regional security and the role of European arms export. SWP Research Paper, 10/2017. Hellmeier, S. (2020). How foreign pressure affects mass mobilization in favor of authoritarian regimes. European Journal of International Relations, 1354066120934527. https://doi.org/10.1177/1354066120934527 Hendrix, C. S. (2010). Measuring state capacity: Theoretical and empirical implications for the study of civil conflict. Journal of Peace Research, 47(3), 273-285. https://doi.org/10.1177/0022343310361838 Houle, C. (2017). Inequality, ethnic diversity, and redistribution. The Journal of Economic Inequality, 15(1), 1- 23. https://doi.org/10.1007/s10888-016-9340-8 Jackson, R. H., & Rosberg, C. G. (1982). Why Africa's Weak States Persist: The Empirical and the Juridical in Statehood. World Politics, 35(1), 1-24. https://doi.org/10.2307/2010277 22 Jervis, R. (1978). Cooperation under the Security Dilemma. World Politics, 30(2), 167-214. https://doi.org/10.2307/2009958 Kollias, C., & Paleologou, S.-M. (2002). Is there a Greek-Turkish arms race? Some further empirical results from causality tests. Defence and Peace Economics, 13(4), 321-328. https://doi.org/10.1080/10242690212357 Kurtz, M. J. (2013). Latin American state building in comparative perspective : social foundations of institutional order. Cambridge University Press. Lee, M. M., & Zhang, N. (2017). Legibility and the Informational Foundations of State Capacity. The Journal of Politics, 79(1), 118-132. https://doi.org/https://doi.org/10.1086/688053 Lindvall, J., & Teorell, J. (2016). State Capacity as Power: A Conceptual Framework (1). Department of Political Science, Lund University. https://portal.research.lu.se/portal/en/publications/state-capacity-as-power-a- conceptual-framework(ca52e85b-f17e-401d-9414-c5f52ec860e9).html Lopes da Silva, D. (2020, 29 October). Autocracy is on the rise: Should we expect military spending to follow? https://sipri.org/commentary/topical-backgrounder/2020/autocracy-rise-should-we-expect- military-spending-follow Lührmann, A., Maerz, S. F., Grahn, S., Alizada, N., Gastaldi, L., Hellmeier, S., . . . Lindberg, S. I. (2020). Autocratization Surges – Resistance Grows (Democracy Report, Issue. Mann, M. (1984). The autonomous power of the state : its origins, mechanisms and results. European Journal of Sociology / Archives Européennes de Sociologie / Europäisches Archiv für Soziologie, 25(2), 185-213. Mansfield, E. D., & Snyder, J. (2007). The Sequencing “Fallacy". Journal of Democracy, 18(3), 5-9. https://doi.org/doi:10.1353/jod.2007.0047 Marshall, M., Gurr, T., & Jaggers, K. (2013). Polity IV Project: Political Regime Characteristics and Transitions, 1800–2012. Meng, A. (2021). Ruling Parties in Authoritarian Regimes: Rethinking Institutional Strength. British Journal of Political Science, 51(2), 526-540. https://doi.org/10.1017/S0007123419000115 Migdal, J. S. (1988). Strong societies and weak states : state-society relations and state capabilities in the Third World. Princeton University Press. Nistotskaya, M., & Cingolani, L. (2015). Bureaucratic Structure, Regulatory Quality, and Entrepreneurship in a Comparative Perspective: Cross-Sectional and Panel Data Evidence. Journal of Public Administration Research and Theory, 26(3), 519-534. https://doi.org/10.1093/jopart/muv026 Papaioannou, K. J., & Van Zanden, J. L. (2015). The dictator effect: how long years in office affect economic development. Journal of Institutional Economics, 11(1), 111-139. https://doi.org/10.1017/S1744137414000356 Partem, M. G. (1983). The Buffer System in International Relations. Journal of Conflict Resolution, 27(1), 3-26. https://doi.org/10.1177/0022002783027001001 Pettersson, T., & Eck, K. (2018). Organized violence, 1989–2017. Journal of Peace Research, 55(4), 535-547. https://doi.org/10.1177/0022343318784101 23 Queralt, D. (2019). War, International Finance, and Fiscal Capacity in the Long Run. International Organization, 73(4), 713-753. https://doi.org/10.1017/S0020818319000250 Richardson, L. F. (1960). Arms and insecurity; a mathematical study of the causes and origins of war. Boxwood Press. Ross, M., & Mahdavi, P. (2015). Oil and Gas Data, 1932-2014 Version V2) Harvard Dataverse. https://doi.org/doi:10.7910/DVN/ZTPW0Y Rothstein, B., & Teorell, J. (2008). What Is Quality of Government? A Theory of Impartial Government Institutions. Governance, 21(2), 165-190. https://doi.org/https://doi.org/10.1111/j.1468- 0491.2008.00391.x Sawani, Y. M. (2018). Public Administration in Libya: Continuity and Change. International Journal of Public Administration, 41(10), 807-819. https://doi.org/10.1080/01900692.2017.1387146 Scott, J. C. (2009). The art of not being governed : an anarchist history of upland Southeast Asia. Yale University Press. Singer, J. D. (1988). Reconstructing the correlates of war dataset on material capabilities of states, 1816– 1985. International Interactions, 14(2), 115-132. https://doi.org/10.1080/03050628808434695 Skocpol, T. (1985). Bringing the State Back In: Strategies of Analysis in Current Research. In P. B. Evans, D. Rueschemeyer, & T. Skocpol (Eds.), Bringing the State Back In (pp. 3-37). Cambridge University Press. Smyth, R., & Kumar Narayan, P. (2009). A Panel Data Analysis of the Military Expenditure-External Debt Nexus: Evidence from Six Middle Eastern Countries. Journal of Peace Research, 46(2), 235-250. https://doi.org/10.1177/0022343308100717 Soifer, H. (2008). State Infrastructural Power: Approaches to Conceptualization and Measurement. Studies in Comparative International Development, 43(3), 231. https://doi.org/10.1007/s12116-008-9028-6 Soifer, H. D. (2015). State building in Latin America. Cambridge University Press. Solmirano, C., & Perlo-Freeman, S. (2009, Jan. 10). Is South America on the brink of an arms race? https://www.sipri.org/commentary/2009/jan-10-south-america-brink-arms-race Straus, S. (2006). The order of genocide : race, power, and war in Rwanda. Cornell University Press. Strøm, K. W., Gates, S., Graham, B. A. T., & Strand, H. (2017). Inclusion, Dispersion, and Constraint: Powersharing in the World’s States, 1975–2010. British Journal of Political Science, 47(1), 165-185. https://doi.org/10.1017/S0007123415000174 Svolik, M. W. (2012). The politics of authoritarian rule. Cambridge University Press. Taylor, B. D., & Botea, R. (2008). Tilly Tally: War-Making and State-Making in the Contemporary Third World1. International Studies Review, 10(1), 27-56. https://doi.org/10.1111/j.1468-2486.2008.00746.x Thies, C. (2007). The Political Economy of State Building in Sub-Saharan Africa. The Journal of Politics, 69(3), 716-731. https://doi.org/https://doi.org/10.1111/j.1468-2508.2007.00570.x 24 Thies, C. G. (2004). State Building, Interstate and Intrastate Rivalry: A Study of Post-Colonial Developing Country Extractive Efforts, 1975–2000. International Studies Quarterly, 48(1), 53-72. https://doi.org/10.1111/j.0020-8833.2004.00291.x Thompson, W. R. (2002). Identifying Rivals and Rivalries in World Politics. International Studies Quarterly, 45(4), 557-586. https://doi.org/10.1111/0020-8833.00214 Tilly, C. (1992). Coercion, capital, and European states, AD 990-1992 (Rev. pbk. ed.). Blackwell. Tilly, C., Ardant, G., & Social Science Research Council. Committee on Comparative Politics. (1975). The Formation of national States in Western Europe. Princeton University Press. von Soest, C., & Grauvogel, J. (2017). Identity, procedures and performance: how authoritarian regimes legitimize their rule. Contemporary Politics, 23(3), 287-305. https://doi.org/10.1080/13569775.2017.1304319 Wang, E. H., & Xu, Y. (2018). Awakening Leviathan: The effect of democracy on state capacity. Research & Politics, 5(2), 2053168018772398. https://doi.org/10.1177/2053168018772398 Yesilyurt, M. E., & Elhorst, J. P. (2017). Impacts of neighboring countries on military expenditures:A dynamic spatial panel approach. Journal of Peace Research, 54(6), 777-790. https://doi.org/10.1177/0022343317707569 Zacher, M. W. (2001). The Territorial Integrity Norm: International Boundaries and the Use of Force. International Organization, 55(2), 215-250. 25 Appendix TABLE A1. SUMMARY STATISTICS Variable Obs Mean Std.Dev. Min Max State capacity 4322 -.366 .79 -3.512 2.22 Neighbours’ military expenditure (log) 4322 14.527 2.472 8.12 19.711 Neighbours’ military expenditure (ordinal sc.) 4322 2.869 1.433 1 5 Executive constraints 4035 2.617 1.648 1 7 Neighbor war history 4322 .321 .467 0 1 GDP p.c. (log) 4322 7.781 1.042 4.889 13.357 Oil wealth (log) 4321 9.56 10.442 0 26.357 Alliance dummy (NATO or Warsaw) 4322 .055 .229 0 1 Military expenditure p.c. (log) 4116 -4.158 2.263 -18.713 3.015 Military personnel p.c. 4274 .007 .008 0 .077 Suffrage 4321 71.575 44.204 0 100 Elections, free & fair 4302 .216 .206 0 .921 Judicial constraints on the executive 4321 .36 .234 .005 .938 Legislative constraints on the executive 3594 .293 .244 .026 .918 Table A1 presents summary statistics for all variables used in the main regression models presented in the paper, as well as all robustness checks. 26 TABLE A2. INDICATORS USED TO CREATE THE CAPACITY-INDEX Admin. & Civil Service (Global Integrity, 2012) Fractal Borders (Alesina et al., 2011) Admin. Efficiency (Adelman and Morris, 1967) (Log) Mountainous Terr. (Fearon and Laitin, 2003) Anocracy (calculated from Polity IV) Political Terror Scale (Gibney et al., 2011) Bureaucratic Quality (Political Risk Services) Quality, Budgetary & Financial Man. (World Bank) Census Frequency (calculated from UN 2011) Quality of Public Administration (World Bank) Civil Service Confidence (World Values Surveys) Rel. Pol Capacity (Arbetman-Rabinowitz et al. 2011) Contract-Intensive Money (WDI) Statistical Capacity (World Bank) Military Personnel per capita (COW) Tax Evasion not Damaging (IMD, 2011) (Log) Military Spending per capita (COW) Taxes on Income as % of Revenue (IMF, WDI) Monopoly Use of Force (Bertelsmann TI) Taxes on Int. Trade as % Revenue (IMF, WDI) Effective Implementation of Gov. Decisions (IMD) Total Tax Revenue as % GDP (IMF, WDI, OECD) Efficiency of Revenue Mobilization (World Bank) Weberianness (Rauch and Evans, 2000) Source: Hanson & Sigman (2013). 27 TABLE A3. COUNTRY-YEARS INCLUDED IN THE SAMPLE Afghanistan 1960-2009 Albania 1960-1991; 1996 Algeria 1960-2009 Angola 1976-2009 Argentina 1962; 1966-1972; 1976-1982 Armenia 1992-2009 Azerbaijan 1992-2009 Bangladesh 1972-1990; 2007-2008 Belarus 1994-2009 Benin 1961-1990 Bhutan 1971-2009 Bolivia 1960-1978; 1980-1981 Brazil 1964-1984 Bulgaria 1960-1989 Burkina Faso 1961-2009 Burundi 1963-2004 Cambodia 1960-2009 Cameroon 1960-2009 Central African Republic 1961-1992; 2003-2009 Chad 1961-2009 Chile 1973-1989 China 1960-2009 Congo 1963-2009 Congo (DRC) 1961-2009 Côte d'Ivoire 1961-2009 Croatia 1992-1999 Cyprus 1975-1976 28 Djibouti 1977-2009 Dominican Republic 1960-1965 Ecuador 1963-1978; 2000-2002 Egypt 1960-2009 El Salvador 1960-1983 Equatorial Guinea 1969-2009 Eritrea 1993-2009 Eswatini 1969-2009 Ethiopia 1960-1992 Gabon 1960-2009 Gambia (the) 1965-1971; 1994-2009 Georgia 1992-2003 Germany, East 1960-1990 Ghana 1960-1969; 1972-1978; 1981-1996 Greece 1967-1973 Guatemala 1963-1965; 1982-1985 Guinea 1960-2009 Guinea-Bissau 1975-1993; 1998-2009 Guyana 1966-1991 Haiti 1960-2009 Honduras 1963-1970; 1972-1981; 2009 Hungary 1960-1989 Indonesia 1960-1998 Iran 1960-2009 Iraq 1960-2009 Jordan 1960-2009 Kazakhstan 1992-2009 Kenya 1964-2001 Korea, North 1960-2009 29 Korea, South 1961-1987 Kuwait 1963-2009 Kyrgyzstan 1992-2009 Laos 1960-2009 Latvia 1992 Lebanon 1960-2009 Lesotho 1967-2001 Liberia 1960-2005 Libya 1960-2009 Malawi 1965-1993 Malaysia 1963-2009 Mali 1961-1991 Mauritania 1961-2009 Mexico 1960-1999 Mongolia 1960-1989 Morocco 1960-2009 Mozambique 1975-2009 Myanmar 1962-2009 Namibia 1990-2009 Nepal 1960-1990; 2002-2007 Nicaragua 1960-1983 Niger 1961-1992; 1996-1998; 2009 Nigeria 1966-1978; 1983-2009 Oman 1971-2009 Pakistan 1977-1987; 1999-2007 Panama 1968-1990 Paraguay 1960-2002 Peru 1962; 1968-1979; 1990-2000 Poland 1960-1988 30 Portugal 1960-1975 Qatar 1972-2009 Vietnam, South 1960-1975 Romania 1960-1990 Russia 1999-2009 Rwanda 1963-2009 Saudi Arabia 1960-2009 Senegal 1961-1999 Sierra Leone 1967-1991; 1993-2001 Somalia 1969-2009 South Africa 1965-1993 Spain 1960-1976 Sudan 1960-1964; 1969-2009 Syria 1961-2009 Tajikistan 1992-2009 Tanzania 1964-2009 Thailand 1960-1974; 1976-1982; 1991; 2006-2009 Togo 1960-2009 Tunisia 1960-2009 Turkey 1960; 1980-1982 Turkmenistan 1992-2009 Uganda 1963-1979; 1985-2009 United Arab Emirates 1972-2009 Uruguay 1973-1984 Uzbekistan 1992-2009 Venezuela 2005-2009 Vietnam 1960-1975 Yemen 1990-2009 Yemen Arab Republic 1962-1989 31 Yemen People's Republic 1968-1989 Yugoslavia 1960-1991 Zambia 1965-2007 Zimbabwe 1970-2009 32 TABLE A4. AUTOCRATIC COUNTRY-YEARS ACCORDING TO GEDDES ET AL. Model 9 Model 10 Neighbours’ military expenditure (log.) 0.112*** 0.0625* (0.0267) (0.0238) Executive constraints -0.294*** (0.0822) Neighbours’ military expenditure (log.) ## Ex- 0.0200** ecutive constraints (0.00596) Constant -4.529*** -3.825*** (0.588) (0.516) Country-fixed effects Yes Yes Year-fixed effects Yes Yes Controls Yes Yes N 3516 3381 Countries 110 110 Time period 1960-2009 1960-2009 Within R2 0.386 0.453 Between R2 0.590 0.567 Overall R2 0.580 0.591 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy). * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. Table A4 displays the results for the first set of robustness tests. Model 9 and 10 replicate model 2 and 6 in the main paper, this time using the autocratic regime data set developed by Geddes et al. (2014) to classify observations as autocratic. 33 TABLE A5. AUTOCRATIC COUNTRY-YEARS = POLYARCHY BELOW 0,5 Model 11 Model 12 Neighbours’ military expenditure (log.) 0.0978*** 0.0679** (0.0270) (0.0222) Executive constraints -0.151** (0.0464) Neighbours’ military expenditure (log.) ## Ex- 0.0106** ecutive constraints (0.00335) Constant -4.256*** -3.629*** (0.498) (0.454) Country-fixed effects Yes Yes Year-fixed effects Yes Yes Controls Yes Yes N 3992 3718 Countries 122 121 Time period 1960-2009 1960-2009 Within R2 0.345 0.435 Between R2 0.620 0.598 Overall R2 0.571 0.599 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy). * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. Table A5 displays the results for the second set of robustness tests. Model 11 and 12 replicate model 2 and 6 in the main paper. Instead of using the dichotomous regime-classification developed by Boix et al. (2012) to classify observations as non-democratic, these regressions are based on all country-years in the 1960-2009 time period where the polyarchy-index (Coppedge et al., 2020) is below 0.5. 34 TABLE A6. AUTOCRATIC COUNTRY-YEARS = POLYARCHY BELOW 0,4 Model 13 Model 14 Neighbours’ military expenditure (log.) 0.101*** 0.0676** (0.0288) (0.0230) Executive constraints -0.169** (0.0551) Neighbours’ military expenditure (log.) ## Ex- 0.0123** ecutive constraints (0.00400) Constant -4.214*** -3.614*** (0.515) (0.464) Country-fixed effects Yes Yes Year-fixed effects Yes Yes Controls Yes Yes N 3642 3387 Countries 116 115 Time period 1960-2009 1960-2009 Within R2 0.347 0.437 Between R2 0.599 0.588 Overall R2 0.570 0.601 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy). * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. Table A6 displays the results for the third set of robustness tests. Model 13 and 14 replicate model 2 and 6 in the main paper. Instead of using the dichotomous regime-classification developed by Boix et al. (2012) to classify observations as non-democratic, these regressions are based on all country-years in the 1960-2009 time period where the polyarchy-index (Coppedge et al., 2020) is below 0.4. 35 TABLE A7. AUTOCRATIC COUNTRY-YEARS = POLYARCHY BELOW 0,3 Model 15 Model 16 Neighbours’ military expenditure (log.) 0.107** 0.0742** (0.0318) (0.0243) Executive constraints -0.202*** (0.0597) Neighbours’ military expenditure (log.) ## Ex- 0.0150*** ecutive constraints (0.00434) Constant -4.171*** -3.527*** (0.553) (0.471) Country-fixed effects Yes Yes Year-fixed effects Yes Yes Controls Yes Yes N 3153 2976 Countries 108 107 Time period 1960-2009 1960-2009 Within R2 0.363 0.454 Between R2 0.587 0.574 Overall R2 0.574 0.596 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy). * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. Table A7 displays the results for the fourth set of robustness tests. Model 15 and 16 replicate model 2 and 6 in the main paper. Instead of using the dichotomous regime-classification developed by Boix et al. (2012) to classify observations as non-democratic, these regressions are based on all country-years in the 1960-2009 time period where the polyarchy-index (Coppedge et al., 2020) is below 0.3. 36 TABLE A8. AUTOCRATIC COUNTRY-YEARS OUTSIDE EUROPE Model 17 Model 18 Neighbours’ military expenditure (log.) 0.0966** 0.0533* (0.0290) (0.0225) Executive constraints -0.228*** (0.0566) Neighbours’ military expenditure (log.) ## Ex- 0.0159*** ecutive constraints (0.00414) Constant -4.542*** -3.683*** (0.510) (0.461) Country-fixed effects Yes Yes Year-fixed effects Yes Yes Controls Yes Yes N 3449 3215 Countries 103 103 Time period 1960-2009 1960-2009 Within R2 0.351 0.448 Between R2 0.610 0.589 Overall R2 0.571 0.605 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy). * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. Table A8 displays the results for the fifth set of robustness tests. Model 17 and 18 replicate model 2 and 6 in the main paper, but excludes all autocratic country-years from Europe. 37 TABLE A9. INTERACT WITH ALLIANCE DUMMY (NATO OR WARSAW-PACT) Model 19 Model 20 Neighbours’ military expenditure (log.) 0.0967*** 0.0555* (0.0284) (0.0224) Alliance-dummy (NATO or Warsaw-pact) 1.787 1.659 (1.210) (1.013) Neighbours’ military expenditure (log.) ## Al- -0.0791 -0.0704 liance-dummy (0.0877) (0.0739) Executive constraints -0.222*** (0.0563) Neighbours’ military expenditure (log.) ## Ex- 0.0152*** ecutive constraints (0.00408) Constant -4.409*** -3.555*** (0.514) (0.465) Country-fixed effects Yes Yes Year-fixed effects Yes Yes Controls Yes Yes N 3736 3498 Countries 117 117 Time period 1960-2009 1960-2009 Within R2 0.340 0.433 Between R2 0.638 0.615 Overall R2 0.587 0.614 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy). * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. 38 Table A9 displays the results for the sixth set of robustness tests. Model 19 and 20 replicate model 2 and 6 in the main paper, but includes an interaction-term between the main independent variable (neigh- bours’ military expenditure) and a dummy which takes the value 1 if a country-year is a member of either NATO or the Warsaw-pact. 39 TABLE A10. CONTROL FOR MILITARY EXPENDITURES PER CAPITA Model 21 Model 22 Neighbours’ military expenditure (log.) 0.0671** 0.0339 (0.0245) (0.0212) Military expenditure p.c. (log.) 0.0948*** 0.0909*** (0.0165) (0.0161) Executive constraints -0.168** (0.0616) Neighbours’ military expenditure (log.) ## Ex- 0.0120** ecutive constraints (0.00427) Constant -2.985*** -2.358*** (0.533) (0.511) Country-fixed effects Yes Yes Year-fixed effects Yes Yes Controls Yes Yes N 3549 3365 Countries 117 117 Time period 1960-2009 1960-2009 Within R2 0.474 0.515 Between R2 0.739 0.721 Overall R2 0.706 0.709 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy). * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. Table A10 displays the results for the seventh set of robustness tests. Model 21 and 22 replicate model 2 and 6 in the main paper, controlling for military expenditures per capita. 40 TABLE A11. CONTROL FOR MILITARY PERSONNEL PER CAPITA Model 23 Model 24 Neighbours’ military expenditure (log.) 0.0894** 0.0521* (0.0293) (0.0231) Military personnel p.c. 18.29*** 12.12*** (4.677) (3.415) Executive constraints -0.240*** (0.0563) Neighbours’ military expenditure (log.) ## Ex- 0.0166*** ecutive constraints (0.00412) Constant -4.240*** -3.421*** (0.513) (0.470) Country-fixed effects Yes Yes Year-fixed effects Yes Yes Controls Yes Yes N 3689 3469 Countries 117 117 Time period 1960-2009 1960-2009 Within R2 0.365 0.439 Between R2 0.704 0.662 Overall R2 0.627 0.646 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy). * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. Table A11 displays the results for the eight set of robustness tests. Model 23 and 24 replicate model 2 and 6 in the main paper, controlling for military personnel per capita. 41 TABLE A12. CONTROL FOR SUFFRAGE & FREE AND FAIR ELECTIONS Model 25 Model 26 Neighbours’ military expenditure (log.) 0.0556* 0.0572* (0.0225) (0.0224) Executive constraints -0.240*** -0.234*** (0.0567) (0.0578) Neighbours’ military expenditure (log.) ## Ex- 0.0161*** 0.0158*** ecutive constraints (0.00412) (0.00417) Suffrage 0.000577* (0.000267) Elections, free & fair 0.0525 (0.101) Constant -3.428*** -3.456*** (0.470) (0.472) Country-fixed effects Yes Yes Year-fixed effects Yes Yes Controls Yes Yes N 3497 3479 Countries 117 117 Time period 1960-2009 1960-2009 Within R2 0.426 0.424 Between R2 0.579 0.588 Overall R2 0.591 0.596 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy). * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. Table A12 displays the results for the ninth set of robustness tests. Model 25 and 26 replicate model 6 in the main paper, controlling separately for suffrage and free and fair elections (Coppedge et al., 2020). 42 TABLE A13. ALTERNATIVE POWER-SHARING MEASURES Model 27 Model 28 Neighbours’ military expenditure (log.) 0.0508 0.0628 (0.0301) (0.0394) Judicial constraints on the executive -1.789* (0.696) Neighbours’ military expenditure (log.) ## Ju- 0.119* dicial constraints (0.0511) Legislative constraints on the executive -1.761* (0.828) Neighbours’ military expenditure (log.) ## 0.0934 Legislative constraints (0.0600) Constant -3.549*** -3.362*** (0.561) (0.627) Country-fixed effects Yes Yes Year-fixed effects Yes Yes Controls Yes Yes N 3735 3044 Countries 117 117 Time period 1960-2009 1960-2009 Within R2 0.341 0.330 Between R2 0.614 0.541 Overall R2 0.566 0.543 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy). * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. 43 Table A13 displays the results for the tenth set of robustness tests. Model 27 and 28 replicate model 6 in the main paper, using two alternative measures of power-sharing: The judicial constraints on the executive index, and the legislative constraints on the executive index (Coppedge et al., 2020). 44 TABLE A14. IDENTIFYING OUTLIERS Executive Neighbours’ military expenditure, Constraints ordinal scale 1 2 3 4 5 Total 1 392 318 261 252 187 1410 2 100 143 128 82 138 591 3 249 238 168 201 299 1155 4 20 2 19 33 40 114 5 25 22 65 28 24 164 6 9 10 8 7 3 37 7 22 29 17 10 1 79 Total 817 762 666 613 692 3550 Table A14 shows the distribution of country-years for different combinations of executive constraints and external military threats. To make the data easier to read, I converted the main independent variable – neighboring countries military expenditures, log transformed – into an ordinal scale, where 1 repre- sents the lowest quintile, 2 the second lowest quintile, and so on. As can be seen, relatively few obser- vations are found in the lower right corner; that is displaying the combination of high executive con- straints and high levels of neighboring countries military spending. This raises the possibility that one of the main results of the paper – that the effect of external military threats on state-building is stronger in more constrained autocracies – are in fact driven by these outliers. To alleviate such concerns, I identify the 73 country-years with high executive constraints (5-7) and high levels of external threats (4-5), displayed in table A15 below. I then rerun the main regressions, either dropping these observations from the sample, or include an outlier dummy as a control (table A16). 45 TABLE A15. COUNTRY-YEARS, OUTLIERS Albania, 1996 Kyrgyzstan, 2006 Nepal, 1990; 2006-2007 Algeria, 2009 Lebanon, 2005-2009 Paraguay, 1992-2002 Armenia, 1992-1994; 1998-2009 Malaysia, 1992-1994; 2008-2009 Romania, 1990 Belarus, 1994 Mexico, 1997-1999 Russia, 2000-2006 Georgia, 1992-2003 Namibia, 1990; 2005-2009 Venezuela, 2005 46 TABLE A16. CONTROLLING FOR OUTLIERS Model 29 Model 30 Neighbours’ military expenditure (log.) 0.0580* 0.0585* (0.0226) (0.0228) Executive constraints -0.225*** -0.223*** (0.0607) (0.0584) Neighbours’ military expenditure (log.) ## Ex- 0.0156*** 0.0151*** ecutive constraints (0.00448) (0.00426) Outlier dummy 0.0947 (0.0880) Constant -3.451*** -3.453*** (0.469) (0.476) Country-fixed effects Yes Yes Year-fixed effects Yes Yes Controls Yes Yes N 3430 3498 Countries 116 117 Time period 1960-2009 1960-2009 Within R2 0.430 0.426 Between R2 0.598 0.579 Overall R2 0.602 0.593 All predictors are lagged one year. Robust standard errors (clustered by country). Controls include GDP p.c. (log), Oil wealth (log) and Neighbor war history (dummy). * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. Table A16 displays the results for the eleventh set of robustness tests. Model 29 and 30 replicate model 6 in the main paper, controlling for outliers. Model 29 excludes the outliers identified in table A14 and A15 from the sample entirely. Model 30 is based on the same sample as model 6 in the main paper, but also includes an outlier dummy as a control. 47