The Joint Effect of International and Domestic-Level State Capacity on Civil War Risk

Abstract

This article examines the role of international and domestic-level factors for strengthening states’ capacity. State failure enhances insecurity, since there is not sufficient agency or institutions to provide adequate security guarantees and to put into operation established rules. When the government is unable to address grievances stemming from such insecurity, armed conflict becomes more likely. Links with external institutions and domestic-level capacity that increases prosperity prevent insurgencies and promote stability, however. To this end, this research develops a new theory linking state capacity and the international and domestic-level factors to internal conflict. Empirically, this study examines the risk of civil conflict onset, focusing on the combined effect of international (political globalisation) and domestic-level (GDP per capita) state capacity as the main driving force. The results show that the joint effect of these factors has a negative impact on civil war risk, which is significantly and substantially important.

1. Introduction

Researchers and practitioners aim to find the origins of intrastate conflict so as to lower the risk of its outbreak and escalation. One of the most well tested arguments on the origins of intrastate conflict refers to the weakness of the state (e.g., Denny and Walter 2014; Tir and Singh 2015). Considering that weak state capacity is one of the key causes of intrastate conflict, it is essential to concentrate on what can strengthen it. This study focuses on the mechanisms that improve state capacity and, thus, decrease civil war risk. Income plays a key role in all economic development and performance theories (Collier 2000; Koubi et al. 2012; Miguel et al. 2004; Murdoch and Sandler 2002). Particularly, civil conflict likely occurs when income is low because rebellions can recruit individuals at low cost (Collier and Hoeffler 2004). Miguel et al. (2004) argue that ‘young men are thought to be more likely to take up arms when income opportunities are worse for them [...] relative to their expected income as a fighter’. Economic constraints increase political turbulence, as they directly impact on human capital, physical capital, and labor growth (Murdoch and Sandler 2002). In turn, this leaves individuals vulnerable, and prone to aggression against a state that is not sustainable (Olson 2008).

While following this line of reasoning, I also argue that state capacity should not only be examined at the domestic level, but also at the international level. Countries are interconnected via many different links, for instance, via diplomatic or trade relations. For example, member-states of the Organisation for Economic Co-operation and Development (OECD) are very well integrated into the global network. Additionally, existing research has challenged the ‘closed polity’ approach to the study of civil war and argues that transnational factors and linkages (e.g., ethnic and regime linkages) between states can exert strong influences on the risk of civil conflict (Forsberg 2016; Gleditsch 2007). To this end, state capacity is enhanced at the international level when a country is well embedded in the international system. A state’s level of political globalisation comprises all those elements that determine whether a country is well connected in the international system, i.e., the number of embassies in a country, participation in UN missions, international treaties, and membership in international organisations (IOs) (Dreher 2006; Dreher et al. 2008). Political globalisation brings with it certain characteristics, such as international recognition, that disfavour the uprising of insurgencies. Dorussen and Ward (2008) add that international links serve as vehicles of communication that promote and preserve peace. As its main contribution, this research studies how international (political globalisation) and domestic-level (GDP per capita) state capacity – jointly – affect the risk of civil conflict, thus addressing a crucial gap on weak governments and civil conflict in the literature.

The rest of this article proceeds as follows. I first introduce the motivation behind the question what the joint effect of the domestic and international factors at enhancing state capacity is. Then, I review some relevant studies on the link between state capacity and civil war and provide a theoretical framework. The following section outlines the research design of the study. The results show that high scores in political globalisation increase the conflict dampening effect of income. Suggestions for further study are discussed in the concluding section.

2. When Government Is the Weakest Link

One of the biggest questions in international politics is why and when conflict occurs. Scholars try to shed light on the origins of civil conflict by identifying the actors and their interests. Another body of the civil wars literature focuses on the reasons why actors choose to fight. For example, Lake and Rothchild (1996) explain that ethnic conflicts are usually caused by collective fears for the future. This atmosphere of uncertainty comes from governments’ incapability to provide a safe environment for the civilians. Along the lines of Fearon and Laitin (2003) there is little evidence that civil war is caused by cultural or ethnic divisions. Instead, the state itself holds responsible for not being able to maintain peace by providing security to the civilians. In this case, literature refers to weak governments that are not in a position to administer the state. Fearon and Laitin (2003) point out that financially, organizationally, and politically weak central governments are mostly at risk of civil war. Specifically they find negative relationship between GDP per capita and civil war because of the deterrent function of state strength. Moreover, Fearon (2004) illustrates the specific characteristics of a weak government. He argues that a government becomes weak after an economic or political shock such as a financial crisis.

Other research refers to political collapse such as sharp change of the political regime as a crucial reason that weakens a state and as a result increases the probability of civil war (DeRouen Jr. et al. 2010; DeRouen Jr. and Sobek 2016; Hartzell and Hoddie 2003). Another contested issue within civil war studies refers to the territorial autonomy and how it affects the strength of a government. The presence of small groups within a state that demand independence or sovereignty creates security concerns and challenges the stability of the government (Coakley 1994; Sisk 1996). This means that a government with limited capacity is unable to preserve its role and legitimacy. The accumulative effects of poverty, unemployment, pressures for territorial autonomy, inefficient educational programs, a poor tax system and scarcity in human personnel are limitations on states’ capability to overcome any security threats from individuals or groups and sustain social rules. For instance, Haiti’s weak capacity to deal with the consequences of the disastrous earthquake in 2010 has potentially prolonged instability in the country.

Hartzell et al. (2001) distinguish two interrelated situations where a state’s weakness is obvious: when a state is subjugated by a single group or coalition of groups and, secondly, when a state is powerless to offer sustained leadership. Under such circumstances, state failure is likely and it raises insecurity because there is no sufficient agency or institution to guarantee and put into operation established rules. Civilians are not satisfied with the government’s policies. When the government is unable to address such situations, then there is a high risk of experiencing civil warfare.

Scholars and practitioners provide various methods that could enhance governments so they are able to stand up against internal threats. For instance, Eizenstat et al. (2005) suggest that sparking economic growth is the best way to prevent state failure. At the same time, Fearon and Laitin (2003) propose that international and nongovernmental organizations should develop programs of support to governments to fight civil war. I combine both lines of argumentation and examine state capacity from two different perspectives in a joint manner: the international and the domestic level.

3. State Capacity at the International and Domestic Level

Existing literature has discussed and defined state capacity in various ways when examining its role in preventing conflict (Balch-Lindsay et al. 2008; Buhaug 2010; Hendrix 2010; Walter 2004), but mostly focused on domestic-level factors. For example, state capacity can be described by military capacity, bureaucratic/administrative capacity, and political institutional coherence and quality of the state (Hendrix 2010). These factors presented by Hendrix (2010) capture very specific governmental sectors. For the purposes of this study, and in an attempt to examine state capacity enhanced by domestic and international factors, I follow the definition given by Eizenstat et al. (2005). The authors delineate that a government should be able to cover several societal aspects for its citizens across three necessary elements: security, legitimacy, and capacity. Along these lines, I explain how state capacity is addressed at the international and domestic level, and potentially decreases the risk of civil war when considering the joint effect of state capacity induced by international and domestic-level factors.

First, a government’s most vital task is to prevent actors from using force threatening the state’s security. This extends to protecting against internal and external fears and preserving sovereignty over territory. If a government is unable to ensure security, rebellious armed groups may use violence to take over the government. The problem here is the so-called ‘security gap’. The ‘security gap’ is first a domestic level factor that can be tackled by imposing the rule of law. To this end, a state needs to secure law enforcement in all areas of policy. For instance, Acemoglu et al. (2014) present three aspects where law enforcement is needed in order to secure stability: human capital, physical capital, and total factor productivity. Securing by law, for instance, obligatory education for all children up to a certain age decreases civil war risk (Rodriguez and Sanchez 2012). What is more, at the international level, institutions and external links such as bilateral or multilateral agreements can provide their members with security mechanisms that protect the sovereignty of the state, including over domestic-level threats (see also Lundgren 2017). Such mechanisms lie on preemtive actions that can legitimate collective decisions and can change perceptions of identity and self-interest (Pevehouse and Russett 2006). For example, some IOs (e.g., NATO) are specialised in military and security affairs. On one hand, such IOs may guarantee their member states that they will support them against (external violent) acts threatening their safety(Bakaki 2016). On the other hand, IOs provide platforms and forums for their members via regular meetings to communicate and exchange ideas over security issues (Bakaki 2018).

Second, a government should come across and be legitimate so as to gain the trust from the society avoiding challenges that could threaten its stability (Eizenstat et al. 2005). For instance, corruption is a common phenomenon that produces mistrust towards a government. Eizenstat et al. (2005) characterize the legitimacy task as risky and controversial. Legitimacy, can be achieved at the domestic level by promoting transparency in decision making, participation, and civil engagement. Information flows are often discussed in the conflict literature as a means of transparency (Beardsley et al. 2017; Eizenstat et al. 2005; Hollyer et al. 2015). Ultimately, a transparent government should provide information how decisions are taken and allow for the engagement of civil society with state institutions. With regards to legitimacy at the international level, political globalisation may support national sovereignty and preserves stability by promoting good governance. Filling the legitimacy gap, a third-party institutionalised actor mostly offers suggestions that usually are based on international recognition and reputation. One of the most common methods is the suggestion of political participation and elections: good offices, facilitation of fair elections, or monitoring of the election processes. External actors and links do have the decorum of legitimate institutions that makes them strong enough to influence the decision of introvert groups or individuals that aim to threaten state security. Third, a government is responsible to provide basic services such as education and health care to its citizens. An inability from the government to do so creates a ‘capacity gap’, which can lead to a loss of public confidence and then possibly political disorder. The ‘capacity gap’ factor at the domestic level is mainly addressed by economic affluence, i.e., the state’s overall financial, administrative, and bureaucratic capabilities. In a peaceful environment, a government is capable of providing public goods to its citizens (Eizenstat et al. 2005; Hegre and Sambanis 2006). But governments may be very much unable to cover citizens’ needs and their well being under other circumstances. When a government is able to provide basic services to its citizens, they are less likely to develop grievances and, as a result, insurgencies are also less likely to occur. The important point against this background is that this factor is not only addressed at the domestic level via the government’s activities, policies, and strengths. It is also addressed via assistance from the outside as other countries or international organisations can help states in need. For instance, international environmental agreements such as the UNFCCC provide financial assistance to promote environmental performance.

Along the aforementioned arguments, I aim to empirically examine whether state capacity impacts on civil war onset. I argue that state capacity should not be solely examined either at the domestic level or the international level, but rather jointly. Although capacity at the domestic level signals a strong state that can anticipate internal threats, it may not be fully sufficient on its own to prevent civil war. At the same time, a state that is well embedded in the international system is also insufficient on its own for strong state capacity. When domestic and international capacity are combined, synergies should develop – one can complement the other.1

4. Research Design

A binary time-series cross-section analysis that uses logistic regression models is used to test the relationship between civil war onset and the joint effect of political globalisation and GDP per capita. This is a global sample that originally comprises all country-years in 1946–2014, but eventually I capture 1970/1971–2014 due to data constraints.

Table 1. Summary statistics.

	Obs.	Mean	Std.Dv.	Min	Max	VIF
Civil war onset	9,620	0.03	0.17	0	1
Political globalisationt−1	6741	54.18	22.87	3.73	98.41	2.64
GDP per capitat−1 (ln)	7105	7.36	1.67	3.62	11.66	2.30
Interaction (joint effect)	6160	445.04	242.96	18.11	1064.95
Democracyt−1	8539	1.22	4.97	−6	7	1.68
Mountainous terrain	8265	2.11	1.43	0	4.56	1.15
Oil resources	8265	0.14	0.34	0	1	1.23
Populationt−1 (ln)	7959	15.79	1.67	11.54021	21.03	1.64
Ethnic fractionalization	8265	0.39	0.28	0.00	0.93	1.26

Dependent variable and interaction term is omitted from VIF.

summarises the descriptive statistics for all variables employed in the analysis as well as the the variation inflation factors (VIFs) of the explanatory variables. Multicollinearity is not an apparent issue since all VIFs are below the common threshold value of 5 (O’brien 2007).2 To minimise bias caused by endogeneity, I temporally lag the explanatory variables by one year. This refers to those indicators that are time variant (i.e., political globalisation, GDP per capita, democracy, and population). Variables that are time invariant are not temporally lagged (i.e., mountainous terrain, oil resources, ethnic fractionalization). Additionally, I employ robust standard errors clustered on country to address potential problems of intra-group correlation.

4.1. Civil War Onset

The dependent variable analysed in this paper concerns civil war onset, and receives a value of 1 if a civil war broke out (0 otherwise) in a given country-year during my sample period. I use the UCDP data based on the following definition: ‘an intrastate conflict is a contested incompatibility that concerns government and/or territory where the use of armed force between two parties, of which at least one is the government of a state, results in at least 25 battle-related deaths’ (Gleditsch et al. 2002; Themnér and Wallensteen 2012). Ongoing conflict years are not included in the onsets coded for my outcome variable. In my sample, 3.26% of the country years have experienced the onset of a civil war between 1946 and 2014 (i.e., 314 observations).

4.2. Main Explanatory Variables

This study theoretically shows that state capacity is not a single-level concept, but we might benefit from considering a joint effect stemming from a combination of levels. To this end, state capacity is influenced by two different levels, i.e., the international and the domestic level. To start with, international state capacity is high if a country is well embedded in the international system (Karreth and Tir 2012). In other words, international connectedness shapes and enhances state capacity. Previously Dorussen and Ward (2008, pp. 196–198) employed a measure of embeddedness to show a state’s involvement in the international system, i.e., the networks generated by sharing memberships in IOs. However, this information is limited because it is derived from the Correlates of War’s IGO data (Pevehouse et al. 2004) that cover only the years until 2005.3 Hence, I employ a ‘political globalisation’ measure to show how well embedded a state is in the international system. The information for political globalisation is taken from Dreher (2006) and Dreher et al. (2008), and I cover the years from 1970 to 2014.4 Political globalisation is a rather inclusive indicator that consists of a series of components that show the connections of a state to the international system: 1 the absolute number of embassies in a country; 2 the absolute number of memberships in international inter-governmental organisations; 3 the number of personnel contributed to the U.N. Security Council Missions per capita; 4 the number of international treaties (documents signed between two or more states and ratified by the highest legislative body of each country since 1945) (Dreher 2006; Dreher et al. 2008). This index, therefore, shows how well a state is embedded in the international system via different institutions. The links with the international system offered via political globalisation can be either direct or indirect. They offer forums of communication, allow and promote information provision, and states have the opportunity to exchange ideas, norms and perceptions.

State capacity at the domestic level is measured in various ways (Hendrix 2010). A series of studies use military capacity of a country to measure strength (e.g., Balch-Lindsay et al. 2008; Buhaug 2010; Henderson and Singer 2000; Walter 2004). Other research measuring state capacity has focused on the administrative capacity of a country, and to what degree a country is able to rule or monitor its population (Fearon and Laitin 2003; Hendrix 2010). Moreover, DeRouen Jr. and Sobek (2004) focus more on the quality of the administration employing a scale from expert assessments. There is, however, an indicator that - to a large degree - captures all these aspects of state capacity and this is economic development (DeRouen Jr. et al. 2010; DeRouen Jr. and Sobek 2016; Fearon and Laitin 2003; Thies 2010). Income addresses and proxies the capacity of a country not only to constrain any rivalries but also to develop and evolve for the benefit of its population (Besley and Persson 2010). Economic development thus influences several aspects of state capacity that previous literature has either examined separately or combined: military, administrative, financial, and bureaucratic capabilities (Balch-Lindsay et al. 2008; Buhaug 2010; DeRouen Jr. and Sobek 2004; Fearon and Laitin 2003; Henderson and Singer 2000; Thies 2010; Walter 2004). Hence, for the domestic-level state capacity, this study employs GDP per capita. In addition, such measure offers a larger spatio-temporal coverage than other measures. I use a lagged value of the natural log of GDP per capita from the World Bank Development Indicators (2017). In my sample, 90% of my sample cases reach a score of at least 88.7 in globalisation, while this variable has a median score of 52.60. Likewise, 90% of country years in my sample reach a score of at least 9.90 in GDP per capita, while the median score is 7.21.

Based on the theoretical discussion of this study, state capacity is driven by the domestic level and the international level. In order to examine state capacity at the international and domestic level jointly, I include a multiplicative interactive term between political globalisation and GDP per capita (ln) (see also Brambor et al. 2005). Note in this context that the two variables are sufficiently different from each other, albeit they are related, and scoring high on one level does not necessarily imply to do well on the other. Consider India, for example, which receives scores above 90 on the political-globalisation measure as of 2010, but has a relatively low income. Conversely, Luxembourg has the sample’s highest income in 2014 (log of GDP per capita of 11.67), but scores only 78 on political globalisation.

4.3. Control Variables

In addition to the main independent variables (i.e., political globalisation, GDP per capita, and their interaction), I expect other determinants to impact civil war onset. When accounting for additional indicators of the dependent variable, I address the issue of omitted variable bias. At the same time, this allows me to largely address the issue of selection bias (selection on observables) since most control variables included in the analysis are also correlated with the main explanatory variables (political globalisation, GDP per capita and their joint effect). Colaresi and Thompson (2002) highlight the necessity for research on conflict onset using variables that can be related to each conflict cross-sectionally. This analysis controls for the level of democracy within each state using the modified polity indicator for democracy from Vreeland (2008). This variable ranges between -6 and +7 in my sample, where negative values approach a purely autocratic regime and positive values pertain to more democratic systems. Another determinant of civil conflict is the population size (Hegre and Sambanis 2006). For Fearon and Laitin (2003), a large population is difficult to control and, as a result, this would increase the number of potential rebels. Information on the population size comes from the World Bank indicators (2017) where ‘total population is based on the de facto definition of population, which counts all residents regardless of legal status or citizenship.

Data from Fearon and Laitin (2003) provide information on ethnolinguistic fractionalization. More ethnically diverse countries tend to be characterised by higher probability of civil conflict onset (Cederman and Girardin 2007; Fearon and Laitin 2003). Mountainous countries and with rough terrain are more likely to experience civil conflicts because the type of the terrain influences the opportunities of rebels for hiding (Fearon and Laitin 2003). For this reason this analysis also controls for the impact of rough terrain on civil conflict using data from Fearon and Laitin (2003). Existing literature on natural resources claims that countries with oil revenues are in a high risk of experiencing a civil conflict. To control for this effect, this analysis includes an indicator for countries in which oil constitutes more than one-third of the total export revenues that equals 1 and 0 otherwise (Fearon and Laitin 2003). Lastly, countries having experienced peace for years are more likely to maintain peace. Hence, I additionally control for the time elapsed since the last intrastate conflict employing cubic polynomials (Carter and Signorino 2007).

5. Empirical Findings

The research’s main models are summarised in

Table 2. Civil war onset and state capacity.

	(Model 1)	(Model 2)	(Model 3)
	Civil War Model	Baseline Model	Full Model
Political globalisationt−1		0.07 ***	0.03 *
		(0.02)	(0.02)
GDP per capitat−1(ln)	−0.26 ***	0.11	−0.07
	(0.07)	(0.18)	(0.20)
Interaction (joint effect)		−0.01 **	−0.01 **
		(0.00)	(0.00)
Democracyt−1	0.01		0.02
	(0.02)		(0.02)
Mountainous terrain	0.08		0.09
	(0.07)		(0.07)
Oil resources	0.70 **		0.61 **
	(0.27)		(0.27)
Populationt−1(ln)	0.29 ***		0.31 ***
	(0.06)		(0.07)
Ethnic fractionalization	1.05 ***		0.99 ***
	(0.31)		(0.31)
Peaceyears	−0.11 ***	−0.16 ***	−0.10 ***
	(0.03)	(0.03)	(0.03)
Peaceyears${}^2$	0.00 **	0.00 **	0.00 ***
	(0.00)	(0.00)	(0.00)
Peaceyears${}^3$	−0.00 **	−0.00 **	−0.00 **
	(0.00)	(0.00)	(0.00)
Constant	−6.49 ***	−3.74 ***	−8.83 ***
	(1.10)	(1.12)	(1.68)
ine N	5,126	5,126	5,126
Pseudolikelihood (ln)	−634.99	−658.25	−633.47
Prob >${\chi }^2$	0.00	0.00	0.00
Area under curve	0.7663	0.7368	0.7671

Standard errors clustered on country in parentheses. * $p<0.10$, ** $p<0.05$, *** $p<0.01$.

, where I examine the relationship between civil war onset and state capacity. Starting with the assessment of the model’s fit, I report logarithmic (pseudo) likelihoods, ${\chi }^2$ test statistics, and the area under the Receiver Operator Characteristic (ROC) curve. The ROC curve statistic is based on an in-sample prediction approach. It theoretically varies between 0.5 (i.e., no predictive power) and 1.0 (i.e., perfect predictive power). Additionally, I employ robust standard errors clustered on country to address potential problems of intra-group correlation and heteroskedasticity. Model 1 depicts a standard civil war model that focuses on the impact of several commonly cited intrastate-related indicators. Model 2 is the baseline model that includes the variables of interest in this study (i.e., political globalisation and GDP per capita (ln) and their interaction). The last model in Table 2 (Model 3) is the full model of my analysis. The last row in Table 2 shows the ROC and area-under-curve (AUC) statistic: the AUC value is 0.767 for Model 3, i.e., the full model. But this value drops to 0.766 in Model 1 that omits the multiplicative specification. Table 2 then already demonstrates that the interaction between GDP per capita and political globalisation is important and should be considered when modelling civil war risk as it has in-sample predictive power. The size and the signs of the coefficients, and the standard errors cannot be directly interpreted. For this reason, I have calculated average marginal effects of GDP per capita (ln) conditional on political globalisation, while holding all other covariates constant at their means (Figure 1 based on Model 3).

When looking at the joint effect of state capacity, i.e., domestic × international level, the results show, in line with my expectations, that as political globalisation increases, the stronger is the conflict-decreasing effect of the GDP per capita (Figure 1). This means that the GDP per capita indicator has a significant impact on reducing the risk of civil war only when political globalisation is well established. In turn, this shows that the two indicators (i.e., political globalisation and GDP per capita) are interrelated, and employing them together makes a crucial difference in reducing the risk of domestic conflict. We should, as a result, expect that a country’s economic development positively interacts with its political globalisation. That said, in low levels of political globalisation, the effect of GDP per capita is insignificant on the risk of civil conflict. Only high levels of participation of a country in the international system generate opportunities for inclusivity, cooperation and development. Otherwise, the country would be isolated and marginalised lacking shared links, norms and ideas offered in the international system. To illustrate the narrative underlying this finding, consider the outbreak of two of the most prominent cases of domestic warfare: on one hand, the Sierra Leone Civil War (1991–2002) and, on the other hand, the Rwandan Civil War (1990–1994). Regarding the former, fighting began when the Revolutionary United Front (RUF) invaded Sierra Leone from Liberia (Abdullah et al. 2004). The dispute “eventually reached all regions of the country and led to major political instability” (Bellows and Miguel 2006, p.394), mostly affecting the civilian population. A few years before the war broke out, in 1985, Major General Joseph Momoh assumed power who was described as a notoriously inept leader (Abdullah et al. 2004, p.93) who further weakened state institutions. However, and in line with the argument, the overall degree of state capacity at the domestic level and internationally was significantly higher in the years before the onset of the civil conflict: on average, Sierra Leone scored 5.36 and 36.33 for domestic and international state capacity, respectively, three years prior to the conflict; in the year fighting broke out, though, my data suggest a value of 5.01 (domestic state capacity) and 33.91 (international state capacity). In other words, state strength at either level dropped in the year of the war outbreak, in fact significantly as my data emphasise a drop by about 6.5–6.6 percent at either level. Regarding the Rwandan Civil War, involving the government of Rwanda and the rebel Rwandan Patriotic Front (RPF) as belligerents (Prunier 1997), there is a similar pattern. With domestic state capacity falling by about 3.8 percent and international state strength dropping by about 0.1 percent in comparison to the year before fighting began, the effect size is not a strongly pronounced as in the case of Sierra Leone, but still suggests that state strength at either level – and jointly considered – are important predictors for the outbreak of domestic armed conflict. That is, while a fully-developed case study is beyond the scope of this large-N study, the narrative from these two illustrations supports the argument developed in this study, namely that low levels of domestic and international state capacity have likely affected – among other factors – the risk of civil war significantly in both cases. This interactive relationship of political globalisation and GDP per capita has a significant impact on the risk of civil war because these two elements jointly affect the capacity of a country at the domestic and also at the international level. Therefore, the international and domestic levels of state capacity should not be treated as substitutes, but rather complements.

Following Ward et al. (2010), I also assess the full model’s predictive power via a 4-fold cross-validation (

Table 3. Out-of-sample prediction: 4 fold cross-validation.

	Full Model	Constrained Model
ine Estimation 1	0.6874	0.6670
Estimation 2	0.7015	0.6776
Estimation 3	0.6995	0.6739
Estimation 4	0.7071	0.6640
Estimation 5	0.6742	0.6705
Estimation 6	0.7034	0.6660
Estimation 7	0.6993	0.6823
Estimation 8	0.7076	0.6712
Estimation 9	0.6963	0.6798
Estimation 10	0.6963	0.6693
ine Mean	0.697	0.672

). This is a quasi-experimental exercise for the full model (Model 3) and the model that omits the interaction terms (Model 1), i.e., GDP per capita (ln), political globalisation and the interaction. To start with, the sample of this analysis has been divided into four parts of roughly the same size. I then use three random parts to measure the parameters, while the fourth part is kept for evaluating the predictive power of either Model 1 or Model 3 on the pooled subparts. I again rely on the AUC-ROC statistic. This process is repeated 10 times for Models 1 and 3, whilst I calculate the average values of the AUC measure across these 10 exercises. Table 3 summarizes the findings and demonstrates that the multiplicative specification also has out-of-sample predictive power: on average, the out-of-sample AUC-ROC decreases by 0.02 percentage points when dropping political globalisation and its interaction with GDP per capita (ln). Therefore, predicting civil war onset is more accurate when we consider the joint impact of domestic and international state capacity.

Figure 2 (based on Model 3) shows the substantive effects for all control variables. It depicts first differences for each variable, i.e., changes in the predicted probability of civil war onset for a variable at its maximum value compared to its minimum value, holding all other variables constant at their respective means. The results, as expected, indicate that population is positively associated with the risk of civil war. For example, civil war risk increases by about 15 percentage points when moving from the minimum to the maximum of Population_t−1(ln). Likewise ethnic fractionalisation has a significant and positive effect on civil war risk, indicating that within country ethnic diversity increases instability and potential unrest. Oil producers, as expected, are more prone to civil war (Fearon and Laitin 2003). The indicators of democracy and mountainous terrain indicate no significant relationship to civil war onset.

To assess the robustness of my results, I also examined various alternative specifications for the main logistic regression model (Model 3), which are discussed in the Supplement Materials. For example, previous research has used other measures to capture state capacity such as the quality of civil services, and the capacity of the state to monitor its population via governmental institutions (Besley and Persson 2009; DeRouen Jr. and Sobek 2004; Hendrix 2010; Sobek 2010; Tilly 2003). For this reason, I employ an alternative measure based on bureaucratic quality (Model 1 in Table S1 and Figure S1 in the Supplement Materials). Countries with a high quality bureaucratic status are most likely to maintain stability and offer effective state services (see also Fukuyama 2013). Although political globalisation is a rather inclusive measure that captures various aspects of embeddiness in the international system, I also provide an analysis employing the alternative measure of IGO network centrality (Model 2 in Table S2 and Figure S2 in the Supplementary Materials).

6. Conclusions

This article sought to build on and extend earlier research on civil war. Existing literature suggests that the international community has had only limited success in managing civil war (Collier et al. 2008; Melander et al. 2016). To this end, more proactive methods should be examined that focus not only on managing and resolving civil war, but rather preventing conflict outbreak in the first place (see also Acemoglu et al. 2014; Karreth and Tir 2012; Tir and Karreth 2018). State capacity enhances a government’s ability to rule and sustain stability, which decreases civil war onset. In this study, political globalisation captures capacity at the international level, while GDP per capita addresses capacity at the domestic level. Domestic and international level capacity have in general been treated as mutually exclusive phenomena and studied in very different ways. Research on state capacity and civil war prevention has been inclined to disregard the transnational dynamics of civil war (Forsberg 2016; Gleditsch 2007) by excluding states’ embeddedness in the international system and ultimately its impact on civil war. By examining the two state-capacity levels’ joint effect, I offer a comprehensive assessment of state capacity.

To this end, this study adds to the existing literature offering a novel perspective on civil war prevention. It points to the joint effect of international and domestic-level state capacity. Most of the existing literature that examines ways of preventing civil wars focuses either on the impact derived from the international system (e.g., IGOs) (Tir and Karreth 2018) or the domestic level, e.g., military capacity, economic growth (DeRouen Jr. and Sobek 2004; Hendrix 2010). The findings of this study illustrate that the interactive relationship of political globalisation and GDP per capita has a significant impact on the risk of civil war. Particularly, the GDP per capita indicator has a significant impact on reducing the risk of civil conflict only when political globalisation is strongly pronounced.

This study not only aims at reassessing the way we treat international linkages and state embeddedness in the global system, but also domestic-level state capacity. From a policy perspective, a state’s progress and strength is jointly driven by domestic level developments and international level inclusivity. The two levels seem to be interrelated and interdependent for better effects against instability and, thus, the risk of civil war. Therefore, instead of dealing too late with state instability (e.g., after civil war’s escalation), a stronger emphasis is necessary on civil war prevention by improving states’ global embeddedness and strengthening their domestic level capacities via economic progress. International organisations, for instance, do offer valuable services like monitoring, enforcement, or financial assistance.

Future research may want to systematically examine the joint effect of the international and domestic levels of state capacity in the context of conflict dynamics. For example, including such measures in models on civil war duration or the transnational dimension of domestic conflict may help us to account for the development of intrastate wars more accurately. Moreover, studying state capacity from various angles is not only important for predicting the onset of civil wars, but also for low-intensity violence and (non-violent) protests, e.g., the 2019 Johannesburg riots or the yellow-vests movement in France. In addition, it seems an effort worth making to study the impact of international and domestic state capacity when examining conflict management and its effectiveness.