Can states buy peace? Social welfare spending and civil conflicts

Transcription

1 Research Articles Can states buy peace? Social welfare spending and civil conflicts journal of peace R ESEARCH Journal of Peace Research 49(2) ª The Author(s) 2012 Reprints and permission: sagepub.co.uk/journalspermissions.nav DOI: / jpr.sagepub.com Zeynep Taydas Department of Political Science, Clemson University Dursun Peksen* Department of Political Science, University of Memphis Abstract This study examines whether the state s ability to provide social welfare services has any major effect on the probability of civil conflict onset. We argue that welfare spending contributes to sustaining peace because the provision of social services reduces grievances by offsetting the effects of poverty and inequality in society. Welfare spending serves as an indication of the commitment of the government to social services and reflects its priorities and dedication to citizens. By enacting welfare policies that improve the living standards of citizens, governments can co-opt the political opposition and decrease the incentives for organizing a rebellion. Utilizing time-series, cross-national data for the period, the results indicate that as the level of the government investment in welfare policies (i.e. education, health, and social security) increases, the likelihood of civil conflict onset declines significantly, controlling for several other covariates of internal conflict. Additional data analysis shows that general public spending and military expenditures are unlikely to increase or decrease the probability of civil unrest. Overall, these findings suggest that certain types of public spending, such as welfare spending, might have a strong pacifying effect on civil conflict, and therefore the state s welfare efforts are vital for the maintenance of peace. Keywords civil war, conflict, social policy, state capacity, welfare spending Introduction Citizen dissatisfaction with existing political, economic or social conditions is common across countries. However, only in some instances does this latent discontent cross a certain threshold and transform itself into a major armed conflict. Why do some states experience political violence in the form of civil conflicts while others do not? Although a considerable amount of research has addressed this question, there is still substantial division among scholars with regard to the causes of civil conflicts (Hegre & Sambanis, 2006). Some studies suggest that grievances and justice-seeking behavior are the origins of civil conflicts; others emphasize the material conditions that favor insurgency. According to the former view, factors that create grievance and a sense of injustice, including relative deprivation, collective disadvantages, and inequality, provide groups with the motivation to use violent means against the state and, hence, carry a great deal of importance in explaining civil wars (e.g. Gurr, 1968, 1970; Regan & Norton, 2005). The opportunity theory, on the other hand, claims that the roots of civil wars are not in politics but in economic and other material conditions. Since participation in insurgency depends on the change in the expected payoff from the insurgency, special attention is devoted to the Corresponding author: ztaydas@clemson.edu * The name ordering is reverse alphabetical and does not denote unequal contribution.

2 274 journal of PEACE RESEARCH 49(2) factors that affect the feasibility of rebellion (e.g. Lichbach, 1990, 1995; Tarrow, 1994; Tilly, 1978; Collier & Hoeffler, 2004; Regan & Norton, 2005). Despite the lack of agreement among scholars as to the major causes of civil wars, there is a widespread consensus in the literature on the importance of state capacity for sustaining peace (Arbetman & Kugler, 1997; Buhaug, 2006; Johnson, 2004; Krasner, 1999; Benson & Kugler, 1998; Fearon & Laitin, 2003). Building on the earlier research that emphasizes the importance of the state, this study examines the effect of an important yet neglected dimension of state capacity: the ability of the state to provide social welfare policies to its citizens. Specifically, we seek to answer the following questions: How does social welfare provision (i.e. education, health, and social security policies) affect the likelihood of civil war onset? Can states decrease dissent in the society through the provision of social welfare services? Although scholars have long suspected a relationship between social spending and civil peace, there has not been any thorough empirical analysis of whether social welfare spending is a significant predictor of civil war onset (Sobek, 2010; Hegre & Sambanis, 2006). 1 Azam (2001: 442) asserts that the occurrence of civil conflict in Africa is intimately related to the failure of government to deliver the type of public expenditure that people want; i.e. with a strong redistributive component such as health and education. Can this argument be relevant for other civil conflicts? We suspect that the lack of cross-national studies on this subject is due mostly to the lack of comprehensive spending data over a long period of time. However, the omission also reflects the need to theorize further about the impact of social welfare policies on civil conflicts. This study tries to fill this lacuna. We argue that social welfare efforts by the state can shape the preferences of citizens in ways that discourage the use of violence for political goals. In addition, higher social spending contributes to peace by improving the living standards of citizens and raising the opportunity cost of insurgency. Using time-series, cross-national data for the period, our empirical findings suggest that as the level of the government investment in welfare policies (i.e. education, health, and social security) increases, the likelihood of civil conflict onset declines significantly, even when controlling for the 1 Welfare effort includes all public spending including social transfers, as well as other policies that further social rights like education and health. Throughout the text, social policy, social welfare policy, welfare policy, and public spending are used interchangeably. effect of other economic, demographic, and political indicators identified by the literature. This finding implies that higher levels of social spending have a strong pacifying effect on civil conflict. The remainder of this study is organized as follows. The next section offers an overview of the literature on the nexus between state capacity and civil conflicts, and points out the limitations of the existing studies on the subject. In the following section, we establish the theoretical linkage between social spending and civil violence. The research design section presents the data and operationalization of the variables, and reports the findings. The final section provides a brief summary of the findings and discusses their implications for civil war research. The relevant literature Civil conflicts are essentially about the control of the state and monopoly over the use of force. State capacity, therefore, holds a central place in understanding the onset and escalation of violent domestic conflicts. There is, however, no agreement in the literature as to what aspects of state weakness make states more likely to face insurgency and how to measure this important multidimensional concept (Arbetman & Kugler, 1997; Sobek, 2010; Buhaug, 2006; Taydas, 2011). 2 Earlier research has examined different dimensions of state capacity, including military, administrative/bureaucratic, political institutional, and economic capacity (Hendrix, 2010; Thies 2010; Taydas, Peksen & James, 2010; de Soysa & Fjelde, 2010; Fjelde & de Soysa, 2009; Englehart, 2009; Suzuki & Krause, 2005; Krause & Suzuki, 2005; de Rouen & Sobek, 2004; Johnson, 2004; Fearon & Laitin, 2003; Henderson & Singer, 2000). One common feature of the existing treatments of state capacity is the emphasis on the coercive (and to some extent the administrative) capacity of the state, which reflects the ability of the state to deter violent challenges to its authority (e.g. Fearon & Laitin, 2003; Henderson & Singer, 2000; Collier & Hoeffler, 1998). In this understanding, state strength is associated with the degree to which it can exert power over citizens. Weak states ones that do not have sufficient bureaucratic penetration into society and efficient institutions of coercion like police and military forces are expected to face numerous difficulties in enforcing laws, imposing 2 For a detailed overview of the state capacity literature, refer to the special issue of the Journal of Peace Research guest-edited by Sobek (2010).

3 Taydas & Peksen 275 order, and maintaining peace among the citizens. They also experience numerous challenges in monitoring, discouraging, and suppressing dissent, which create conditions that are conducive for insurgents to recruit noncombatants to their forces. Azam (2001: 435), on the other hand, maintains that the ability to commit credibly is the relevant basis for distinguishing between a strong and a weak state. A weak state, which is unable to commit credibly, will impose too much repression and not enough redistribution. Similarly, Levi (2006: 5) suggests that a strong state is one which is (a) representative and accountable to its people and (b) effective, which means capable of protecting the population from violence and providing other public goods and services that the populace demands. She claims that the state s ability to secure the cooperation and compliance of its citizens is critical for preventing the emergence of violent predation. In this study, following Azam and Levi s logic, we emphasize the state s ability to provide welfare services and argue that the way the state spends its resources is of critical importance to political stability and maintenance of civil peace. We posit that focusing only on the coercive capacity of the state leads to a limited understanding of the central role the state plays in the emergence of internal conflicts. Beyond deterring insurgency and providing security against violence, the state needs to maintain its legitimacy and elicit citizen support. Social spending is a powerful instrument that enables the state to achieve these goals. By enacting welfare policies that have direct and positive impacts on the living standards of citizens, governments can outspend the opposition, which can form the backbone of insurgency. This will help to gain the quasi-voluntary compliance of a broad segment of the population, co-opt political opposition, and decrease the incentives for organizing a rebellion (Levi, 2006). Therefore, we do not treat the state as a passive actor that mainly reacts to insurgency through repressive means. Such a view underestimates the interactions between the state and citizens as well as the role the state plays in the development of grievances. More importantly, treating the state as a passive actor does not account for a state s preventive or strategic actions to secure the compliance of social groups and discourage the emergence of largescale violence before it occurs. In sum, we perceive the state to be an active participant that makes strategic calculations and uses its power in ways that contribute to peace (Burgoon, 2006; Azam, 2001; Bueno de Mesquita et al., 2003; Levi, 2006). Moreover, we seek to tap into state capacity by using a more nuanced and direct measure: welfare spending that accounts for total government expenditures on education, health, and social security services. This measure specifically captures the overall welfare effort of the state and accounts for government spending with a strong redistributive component (Azam, 2001). It therefore serves as an indication of the political commitment of the government to social services and reflects its priorities and dedication to its citizens. 3 We particularly focus on welfare spending because social welfare policies aim to meet the needs of the population (such as education, health, social security, etc.) and improve their social conditions. These policies have immediate, direct, and stronger implications on the citizen s daily lives than other forms of public investments such as military spending and state subsidies for domestic industries. By providing public services and creating social safety nets, states can prevent disadvantaged members of society from falling below a certain level of poverty and experiencing absolute desperation. These conditions might pose a significant threat to the stability of the system and status quo. In addition, the provision of social services has direct implications for the legitimacy and credibility of the state and the potential to raise the cost of rebel labor. Simply put, by spending on public goods and offering benefits in return for citizen loyalty, governments can co-opt political opposition and reduce incentives to join rebel movements. Social welfare policies and civil wars In his study of internal conflicts in Africa, Azam argues that it takes some conscious efforts by the government for a state of peace to be maintained, with some clear impact on public finances (Azam, 1995: 173; 2001). He claims that a strong and credible state manages to devote a large share of its revenues to redistributive policies and failure to accomplish this fundamental task might trigger political violence. We suspect that the relationship between the state s ability to deliver public expenditure that directly benefits the citizens and the emergence of civil conflict can also be applied to other parts of the world (Azam & Mesnard, 2003). The state s 3 The impact of one particular type of policy, such as public health or primary/secondary education, on individuals decision to take part in rebellious activity has been examined by other scholars (see for example Thyne, 2006). We take a different approach, however, and suggest that what matters for civil peace is the overall welfare effort of the state and aggregate measures. We do not have any strong theoretical reason to expect one type of social policy to decrease the risk of civil conflict in the absence of other welfare policies.

4 276 journal of PEACE RESEARCH 49(2) welfare effort might influence the risk of civil conflict in a number of ways, but two in particular appear to be the most crucial factors. First, the government s commitment to redistribute has an important impact on the perceptions and preferences of social actors towards the state. In return for productive social welfare policies, political leaders gain public loyalty, compliance, and support; they also have an easier time pursuing their agendas. This strategy, namely government commitment to redistributing in favor of the poor, was used by rulers in the West in the 19th century to increase the legitimacy of governments and to prevent revolution (Acemoglu & Robinson, 2000). The second and more indirect connection emphasizes the role of social spending in promoting economic development, decreasing the impact of poverty, and undermining the opportunity structure for rebellion. The following section discusses each of the two mechanisms in detail. Modern states have a variety of tasks such as enforcing laws, extracting taxes, ensuring order, determining rules, dealing with emergencies, and providing security as well as other public goods and the capacity of the state is an important determinant of its success. The government s performance, policies, and decisions, in turn, determine its legitimacy and credibility. Legitimate and capable governments are regarded as worthy of support by citizens and can therefore induce compliance without resorting to the use of physical force (Eckstein, 1971; Jackman, 1993). They also have higher chances of maintaining their position in power because societal actors are more likely to support preferences of the state and its designated officials (Nordlinger, 1987). On the other hand, the failure of the state to perform its fundamental tasks can hurt the credibility of the government and decrease its chances of staying in power and, more importantly, can lead to political violence (Azam, 2001). One of the primary duties of the government is to improve the social welfare of citizens, and public spending is an important tool that allows governments to peacefully redistribute and transfer some resources to the public. By investing in social safety nets, in-kind transfers, and valuable goods that are underprovided by the private sector and would not otherwise be available to certain segments of the society, government intervention can have a positive impact on citizens livelihoods and prove that the state cares about its citizens (Van de Walle, 1996; Thyne, 2006). Social policies such as health, education, and social security might also provide direct social insurance and make an immediate, direct impact on people s conditions (Snyder & Bhavnani, 2005; Burgoon, 2006). These kinds of welfare policies can mitigate the effects of poverty, lower societal grievances originating from absolute and relative deprivation, improve the conditions of the poor, and decrease socioeconomic inequality (Van de Walle, 1996; Burgoon, 2006). 4 As the earlier research suggests poverty can certainly make a person outraged and desperate, and a sense of injustice, related particularly to gross inequality, can be a good ground for rebellion even bloody rebellion (Sen, 2008: 8; Gurr, 1968). 5 Bueno de Mesquita et al. (2003) argue that government spending decisions are of primary importance for the political survival of governments and the stability of the system. Specifically, given that citizen attitudes toward the state are informed by self interest, the redistribution of state resources and allocation of a higher percentage of revenues to projects that benefit a large segment of society can enhance the ties between citizens and the state and, more importantly, enhance the power base of the regime. Higher government social spending shows citizens that the government cares about their constituency and that it has an interest in improving their lives. Through welfare policies, governments can affect the attitude and voting choice of the poor, who would otherwise oppose the government. When a great deal of money is spent on economically and socially unproductive areas such as rent-seeking activities only a small portion of the revenue will be 4 Welfare policies might affect the likelihood of civil conflicts through their impact on the levels of poverty, economic insecurity, and inequality in society. Although a quantitative test of each causal mechanism is beyond the scope of this article, we cite several examples of quantitative research and case studies to substantiate our theoretical argument. We attempt to show here that welfare services, on balance, have a positive influence on the conditions and perceptions of citizens, especially those who need these services the most. This, in turn, will diminish people s motivation and ability to engage in violent internal dissent. 5 Despite the strong theoretical reasoning behind the inequality and civil conflict relationship (Gurr, 1970; Regan & Norton, 2005), the existing evidence on this topic is neither consistent nor conclusive (Thoms & Ron, 2007). Most of the influential large-n studies conclude that there is no obvious regularity in the interaction between inequality and civil conflict (Collier & Hoeffler, 1998, 2004; Fearon & Laitin, 2003; Hauge & Ellingsen, 1998; Moore, Lindstrom & O Regan, 1996; Weede, 1987). Many attribute the lack of empirical support for the inequality civil conflict nexus to the use of crossnational, aggregate, class-based indicators that fail to capture both the geographic variation within states and inequalities at the local and group levels. In the absence of more specific and high-quality inequality indicators measuring different types of inequality that citizens experience, it is inappropriate to dismiss ideas about inequality as a possible determinant of civil conflict (Østby, 2008).

5 Taydas & Peksen 277 available for vital redistributive and development policies. Under such conditions, the allegiance that the government enjoys will be limited to those who are favored and rewarded by the government thorough provision of private goods, and hence governments will not be able to attract support from a large constituency (Bueno de Mesquita et al., 2003). Moreover, the legitimacy of the state will suffer and government will foment grievances (Snyder & Bhavnani, 2005). This is especially true for the poorest and most vulnerable segments of society, who are heavily dependent on public services for their survival and cannot substitute public services, like health care and schools, with private ones (Gupta, Verhoeven & Tiongson, 2003; Nouriddin & Simmons, 2006). The under-provision of public services or total neglect of certain communities (i.e. selective provision of goods) can generate a desperate sense of human insecurity and encourage the deprived to take up arms against the government (Burgoon, 2006; Gill & Lundsgaarde, 2004; Azam, 2001). In addition to the immediate impact of social policies on violence, the provision of welfare policies also has various indirect effects on mitigating the risk of civil wars. The most obvious indirect connection between social policy and civil war is the impact of welfare spending on economic wealth. Though scholars differ on why the wealthierstatesareatlowerriskofcivilwars,thereisalmost unanimous agreement about the positive relationship between economic prosperity and civil peace. Earlier cross-national research indicates that countries with higher social expenditure levels especially on education, health care, and social security experience lower levels of poverty (Granato, Inglehart & Leblang, 1996; Adeola, 1996; Gupta, Verhoeven & Tiongson, 2003; Lanjouw et al., 2001; Moon & Dixon, 1992). Even though the type of welfare policies as well as the group they target and the direct impact they have on different segments of the society shows some variation, the well-established research in this area points out that government spending on welfare facilitates long-term productivity and economic development. It consequently undermines poverty and promotes economic prosperity (Benabou, 1997; Dollar & Kraay, 2001; Lindert, 2004; King, 1998; Justino, 2003). Furthermore, government investment in human capital has a direct and positive influence on social mobility, labor opportunities, and the distribution of earning capacities of individuals and households (Korpi & Palme, 1998; Justino, 2003; Rudra, 2004; Chu, Davoodi & Gupta, 2000; Thyne, 2006). It leads to more competitive, effective, productive and skilled workers, which enhance the international competitiveness of the economy (Schultz, 1963; Van de Walle, 1996; Avelino, Brown & Hunter, 2005; Brown & Hunter, 2004; World Bank, 1991, 1993; King, 1998; Burgoon, 2006; Gupta & Verhoeven, 2001). Moreover, two key elements of social welfare education and health have positive externalities for the entire society (Lanjouw et al., 2001; Brown & Hunter, 2004; Azam, 2001). Lanjouw et al. (2001: 18 19), for instance, suggest that the reason governments promote higher standards in health and education is because of their intrinsic value in raising capabilities and individual freedoms. They also have an instrumental value in contributing to higher incomes and in reinforcing each other. The main asset of the poor is their labor. Education and health are critical to preserving and enhancing the quality of this asset and for this reason investment in health and education is especially important for the poor. Furthermore, the effect of health and education is not limited to those who receive the service; it can benefit others who live close to those who receive the investment or the second generation children of the recipients of such social services. Therefore, one effective way of undermining the opportunity structure for rebellion is to promote government investment in social policies. As social policies improve the quality of life in society and promote wealth, the likelihood of domestic violence will decrease. As well as the positive impact of public spending on improving the living standards of citizens, higher social welfare spending is also associated with lower economic inequalities (Sen, 1973; Korpi & Palme, 1998; Burgoon, 2006). That is, an increase in public spending for education, health, and other redistributive welfare policies will allow different segments of society, especially the poor, to have access to these services. Gupta, Verhoeven & Tiongson (2003) and Bidani & Ravallion (1997) claim that public health spending matters more to the poor and public spending in welfare policies often affects the poor more favorably than the non-poor. For example, Gupta, Verhoeven & Tiongson (2003: 694) find that a 1% increase in public spending on health reduces child mortality by twice as many deaths among the poor. Similarly, Le Grand (1987) finds a negative relationship between health inequality and the share of public spending on health care. In sum, social safety nets and the availability of social services are crucial for the lower strata. As scarce resources are allocated for public benefit, this will reduce vulnerability of the poor and contribute to the maintenance of civil peace (Kaufman & Segura-Ubiergo, 2001; Burgoon, 2006). All in all, we suggest that the state can shape the preferences of citizens in ways that discourage the use of

6 278 journal of PEACE RESEARCH 49(2) violence for political goals. In this process the state s ability to provide social services with a strong redistributive component is critical for the maintenance of peace. We therefore hypothesize that the higher the level of government investment in social welfare policies, the lower the risk of civil conflicts. Research design In the preceding section, we argue that higher levels of social welfare spending by the government will reduce the likelihood of civil conflict onset. To quantitatively examine this hypothesis we use time-series, cross-section data for a sample of 153 countries for the period, inclusive. Dependent variable: Civil conflict onset To predict the impact of welfare spending on the initiation of civil conflicts, we use the list of intrastate armed conflicts collected by the UCDP/PRIO Armed Conflict Dataset (Gleditsch et al., 2002). An armed conflict is defined as a contested incompatibility that concerns government or territory or both where the use of armed force between two parties results in at least 25 battlerelated deaths. Of these two parties, at least one is the government of a state (Gleditsch et al., 2002: ). Following common practice in the literature, the Onset of civil conflict variable is operationalized as a dichotomous variable, coded 1 for the year that a civil conflict starts and 0 otherwise. This measure accounts for both low-scaled armed conflicts and fullblown major civil conflicts. To examine whether social welfare spending has a differential impact on major civil conflicts (defined as armed conflicts with more than 1000 battle-related deaths), we also use the Onset of major civil conflict as the second outcome variable. This variable is coded 1 for each country-year in which a major civil conflict with more than 1000 battlerelated deaths breaks out and 0 otherwise. Independent variable: Welfare spending To analyze the effect of welfare spending on civil wars, we use the Welfare spending variable, which is measured as a percentage of GDP. This variable directly accounts for welfare spending in three key areas including education, health, and social security. These are the welfare areas in which the government is expected to play a major role in satisfying the needs of citizens. Health expenditures refer to the basic and hospital health care expenditures. Education expenditures incorporate all the government investments and expenditures in public education. Social security expenditures refer to such expenses as public pensions and unemployment benefits. Our conceptualization of public spending serves as an indicator of governments political commitment. Though there is no guarantee that spending money on certain welfare areas will automatically lead to improvements in that area, the fiscal share of public expenditures reflects priorities set by the government within the public sector. The allocation of greater resources increases the potential for improvements in social services (Brown & Hunter, 2004). Since we are interested in the overall magnitude of governments fiscal commitments to various social needs, we employ an aggregate measure that includes multiple components capturing somewhat different welfare objectives (i.e. health, education, and social security). We gathered the data from three different sources including Kugler, Feng & Zak (2002), Burgoon (2006), and World Bank (2009). In the dataset, Somalia has the lowest welfare spending of 0.7% of GDP in 1986 and Finland has the highest spending with 35.2% in Missing observations in the welfare spending variable, especially in developing countries, is an important issue that may pose difficulties in interpreting results. To address this issue, we use Stata s ICE imputation procedure. While we do not report here to save space, we also repeated the analysis by creating m completed datasets (in our case m ¼ 5) using the Amelia II (King et al., 2001) multiple imputation program. When we used the alternative program, there was no major change in the results reported below (see the online appendix). The imputation procedure specifically imputes missing observations based on the observed values of the non-missing variables in the same row and creates a completed dataset. In the new completed dataset the observed values are the same, while the missing observations are filled in with the imputed data. We then took the average of the spending data gathered from the imputed datasets to include in the models reported below. We also tried to use the modified zero-order regression as an alternative method (Greene, 2003). Specifically, we replaced missing values of the spending variable with zeros and included a dummy variable indicating missing observations to the specification. The modified spending variable was statistically significant in the model. But, because the dummy variable for the missing observations was statistically significant, this method is not appropriate for dealing with the missing observations for welfare spending. Listwise deletion,

7 Taydas & Peksen 279 which simply excludes the missing observations from analysis, is another possible strategy for handling missing data. Because the welfare data are mostly missing for developing countries, which reflects the non-random nature of the missing data, listwise deletion is not appropriate for our analysis (King et al., 2001). Control variables Since there is no common set of control variables used in the literature, we include four of the most widely used and robust controls in the quantitative civil war research in our models. The first control variable is the natural log of GDP per capita (in 1995 constant US dollars) and it is included to account for the assertion that countries with high levels of economic development are less likely to experience domestic violence (e.g. Fearon & Laitin, 2003; Collier & Hoeffler, 2004; Ellingsen, 2000; Barbieri & Reuveny, 2005). To control for the impact of regime type on civil wars, we use the Polity index. The Democracy variable takes the value of 1 for those countries that have a score of 6 or higher on the democracy autocracy index. Anocracy (inconsistent regimes) takes the value of 1 if a country s score falls between 5 and 5 on the 21-point (from 10 to 10) polity scale. Mixed political systems are expected to be more susceptible to civil conflict than pure autocracies and democracies because these countries are partly open yet somewhat repressive (Hegre et al., 2001: 33; Thyne, 2006; Fjelde & de Soysa, 2009). Democracies, on the other hand, are expected to experience a lower risk of civil conflict given that democratic governments are more accountable and responsive and less repressive. 6 The log of total Population is used to control for the expected positive association between higher population rates and the likelihood of civil wars. To check the robustness of the results to the inclusion of some other independent factors, we add three other variables to the model including GDP growth, Oil exporter, and Ethnic fractionalization. GDP growth is the yearly growth rate of GDP per capita. The expectation is that as the level of industrialization increases, the level of economic prosperity will likely increase and socioeconomic dissatisfaction in the society will decrease. The Oil exporter variable is a dichotomous measure that takes 6 Vreeland (2008) criticizes the use of the Polity2 index variable because of the endogeneity of certain components of this measure to civil conflicts. There is no change in the main results when only the non-endogenous components of the Polity2 index are included in the model. the value of 1 if a country receives more than one-third of its export revenues from oil exports, and 0 otherwise. The extant literature considers the effect of natural resources because these resources can be used to finance rebellion (Fearon & Laitin, 2003). The Ethnic fractionalization variable controls for the suggested positive relationship between ethno-linguistic diversity and violence (Ellingsen, 2000; Montalvo & Reynal-Querol, 2005). It varies from 0 (total homogeneity) to 1 (total heterogeneity). The data for ethnic fractionalization and population are from Fearon & Laitin (2003). Methodology Before reporting the findings from the data analysis, there are a few methodological issues that require attention. First, since our dependent variable is a dichotomous measure, logit regression is selected as the method of estimation. Second, all models are estimated using Huber- White corrected robust standard errors (clustered on countries) to deal with the heteroscedasticity problem. Third, to reduce simultaneity bias and make sure that our variables precede the dependent variable, we lag the righthand side variables, except those that are time invariant. Fourth and finally, following earlier practices (e.g. Hegre et al., 2001; Buhaug, 2006), we also create a decay function of the length of time since the end of a previous civil war or the year of independence to control for temporal dependence in the models. This control variable is given by 10 to the power of (number of years since the occurrence of civil war or independence / a), where a represents the half-life parameter. Findings Table I analyzes the relationship between welfare spending and civil conflict onset. 7 The outcome variable in the first four models is the onset of civil conflict variable that accounts for both minor and major civil conflicts. We repeat the analysis in the next four models (Models 5 8) using the onset of major civil conflict as the outcome variable. According to the results in Model 1, the coefficient of welfare spending is negative and statistically significant, indicating that welfare spending is a significant predictor of civil conflict onset, controlling for the four other major predictors of civil conflicts. To check the robustness of the results to the inclusion of other key covariates of civil conflicts, we add GDP 7 Diagnostic tests reveal that there was no issue with multicollinearity in any of the estimations.

8 280 journal of PEACE RESEARCH 49(2) growth, Oil exporter, and Ethnic fractionalization in Models 2, 3, and 4 of Table I, respectively. The results across the models suggest that the inclusion or exclusion of these three additional controls has no major impact on the statistical association between welfare spending and civil conflicts. 8 In Model 5, we repeat the analysis using the onset of major conflicts as the outcome variable. According to the results in the model, welfare spending is statistically significant in the expected direction. When we include the three additional control variables in the models individually (Models 6 8), the spending variable remains statistically significant. Overall, these results in Table I grant convincing empirical support for the argument that the state might buy peace and discourage the use of violent methods via strong social welfare provision. 9 The results for the control variables in Table I are largely consistent with previous research. The population variable shows a statistically significant and negative relationship with civil conflicts in all of the models. The decay function variable is also significant across the models, supporting the assertion that past conflict increases the risk of recurring conflict in the future. The level of economic development appears to be an important predictoroftheonsetofcivilconflicts.oncewerunthe model using only major civil conflicts, however, the suggested impact of economic development on civil conflicts weakens considerably. The results show that anocracy is another significant covariate of major civil conflicts. Among the three control variables added to the analysis, the oil exporter variable is significant for both outcome variables, while ethnic fractionalization 8 To make sure that the inclusion of economically developed countries in the analysis does not bias the results, we estimated the models on a subsample of non-oecd countries. We found that changing the universe of cases does not lead to any substantial difference in our results. We also restricted the data analysis to the post- Cold War (post-1989) era to examine if the decline of East West related ideological confrontation has influenced the relationship between civil conflicts and welfare spending. In these models, welfare spending remained a significant covariate of civil conflict onset. In addition, we checked the robustness of our findings to the inclusion of several other independent variables, including economic openness (trade as a percentage of GDP), corruption in government, bureaucratic quality, regional dummies, official development assistance, and a Cold War dummy. None of these additional controls had a major impact on the results for welfare spending. The results of additional robustness tests are available in the online appendix. 9 The use of Fearon & Laitin s (2003) civil war onset measure as an outcome variable does not lead to any significant change in our results. is significant only in Model 4 that uses the onset of all conflicts as the dependent variable. In Table II, we examine the substantive impact of welfare spending and other significant control variables in the first four models for the onset of all conflicts in Table I. Specifically, we examine the extent of the change in the predicted probability of civil conflict onset once we increase the value of the variables by one or two standard deviations, holding all other independent variables at their mean values. 10 The results suggest that a onestandard deviation increase in welfare spending (as a percentage of GDP) decreases the predicted probability of civil conflict by 23%. When there is a two-standard deviation increase in the welfare spending variable, the probability of civil conflict onset decreases by 41%. How well does the welfare spending variable perform in comparison with the GDP per capita, Population, Oil exporter, and Ethnic fractionalization measures? As the GDP per capita score increases by one standard deviation, the predicted probability of civil conflict decreases by 18%. In the case of a two-standard deviation increase in the same measure, the probability of civil conflict goes down by 36%. Based on these results, it appears that higher public investment in welfare policies could be more beneficial than higher levels of economic development; an increase in the value of welfare spending performs about 5% better in decreasing the predicted probability of civil conflicts than the same level increase in the score of GDP per capita. This suggests that while economic prosperity is essential to improve the living standards of citizens, the extent of wealth redistribution through public spending is crucial to satisfy basic needs and undermine violent insurgencies. According to Table II, a one-standard deviation increase in the population measure leads to a 29% increase, while a two-standard deviation increase leads to a 43% increase in the predicted probability of civil conflict. A one-standard deviation increase in the ethnic fractionalization measure, on the other hand, leads to a 41% increase, while a two-standard deviation increase results in a 91% increase in the predicted probability of civil conflicts. Finally, when we shift the oil exporter variable from 0 to 1, the predicted probability of civil conflicts increases by 95%. Overall, these results indicate that population size, the level of ethnic fractionalization, and oil wealth are likely to have larger substantive impact 10 SPost Stata ado files by Long & Freese (2001) are used for the post-estimation interpretation of regression models for categorical outcomes.

9 Table I. Public spending and civil conflict onset All conflicts Major conflicts Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Welfare spending 0.066** (0.033) 0.067** (0.033) 0.057* (0.034) 0.065** (0.033) 0.096** (0.043) 0.100** (0.042) 0.083* (0.046) 0.095** (0.043) GDP per capitay 0.180** (0.090) 0.189** (0.092) 0.260*** (0.091) (0.091) (0.124) (0.127) 0.275* (0.159) (0.139) Populationy 0.264*** (0.101) 0.287*** (0.100) 0.261*** (0.097) 0.274*** (0.086) 0.294*** (0.107) 0.310*** (0.116) 0.252** (0.099) 0.284*** (0.106) Democracy (0.407) (0.405) (0.376) (0.363) (0.519) (0.519) (0.501) (0.500) Anocracy (0.260) (0.260) (0.257) (0.267) 0.684* (0.353) 0.655* (0.355) 0.784** (0.342) 0.654* (0.345) GDP growth (0.016) (0.032) Oil exporter 0.779*** (0.269) 1.055*** (0.390) Ethnic fract *** (0.434) (0.664) Decay function 0.620** (0.294) 0.575** (0.294) 0.580** (0.278) 0.508* (0.264) 0.899** (0.385) 0.849** (0.368) 0.742* (0.395) 0.818** (0.396) Constant 6.122*** (1.615) 6.412*** (1.629) 5.851*** (1.571) 7.721*** (1.563) 8.206*** (2.218) 8.453*** (2.392) 6.910*** (2.105) 8.662*** (2.328) R-squared Observations Robust standard errors adjusted for clustering over country appear in parentheses. All time-variant independent variables are lagged one year. *** Significant at 1%, ** at 5%, * at 10%. y Logged variables.

10 282 journal of PEACE RESEARCH 49(2) Table II. Predicted probability of civil conflict onset Independent variables Mean Mean þ s Abs. change (% change) Mean þ 2s Abs. change (% change) Welfare spending ( 23) ( 41) GDP per capita ( 18) ( 36) Population (29) (43) Oil exporter* (95) Ethnic frac (41) (91) * For the Oil exporter dummy variable, only the result for the change from minimum (0) to maximum (1) value of this variable is reported. Table III. Total public spending, military expenditures, and civil conflict onset All conflicts Major conflicts Model 1 Model 2 Model 3 Model 4 Total public spending (0.013) (0.049) Military spendingy (0.118) (0.187) Welfare spending 0.073** (0.034) 0.122*** (0.046) GDP per capitay 0.283*** (0.102) 0.233* (0.140) (0.124) (0.188) Populationy 0.331*** (0.111) (0.153) 0.354*** (0.130) (0.203) Democracy (0.394) (0.415) 0.885* (0.488) (0.576) Anocracy (0.274) (0.279) (0.410) 0.735* (0.412) Decay function 0.654** (0.318) 0.546* (0.317) 0.896** (0.442) (0.411) Constant 7.321*** (1.712) 5.740*** (2.032) 9.101*** (2.842) 6.052** (2.701) R-squared Observations Robust standard errors adjusted for clustering over country appear in parentheses. All time-variant independent variables are lagged one year. *** Significant at 1%, ** at 5%, * at 10%. y Logged variables. on the predicted probability of civil conflict onset than welfare spending; yet, the impact of welfare spending is far from negligible. Figure 1 demonstrates the change in the predicted probability of civil conflict onset, when the welfare spending variable is amended from its minimum to the maximum value (in Model 1). To undermine the impact of outliers in the welfare data, in Figure I we use 2% as the minimum and 25% as the maximum scores. As is clearly shown by the figure, an increase in the average amount of welfare spending of about 23 percentage points (from 2% to 25%) reduces the predicted probability of civil conflict onset by more than 70% (from to 0.010). When the average welfare spending score is higher than 15%, the substantive impact of this variable decreases considerably. This suggests that a positive change in welfare spending is crucial, especially for countries that spend less than 15% of their GDP on the provision of social policies. Not surprisingly, when we examine the raw data for welfare spending there are several countries that spend, on average, less than 10% of their GDP on the provision of social policies and have experienced civil conflicts, such as Algeria, Bangladesh, Haiti, Peru, Sri Lanka, Turkey, Uganda, and Zimbabwe. Extending the analysis: General public spending, military expenditure, and civil conflict onset Earlier research on public expenditure suggests that general government expenditure might also have a peace-inducing impact and that higher levels of government spending decrease the risk of civil conflicts (see also Acemoglu, 2005; Fjelde & de Soysa, 2009). This expectation originates from the fact that total government consumption is the broadest measure of government intervention and contributes to economic development. Burgoon (2006: 187) suggests that although some public expenditures do not serve redistributive or welfare goals (e.g. defense spending), they still affect the social rights of citizens, particularly in developing country settings where more explicitly social welfare is not developed. Since it has direct relevance to our research question, in Table III we examine the impact of general public

11 Taydas & Peksen 283 Probability of Civil Conflict Onset 95% upper limit 95% lower limit Welfare Spending (% of GDP) Figure 1. Welfare spending and predicted probability of civil conflict onset spending and military expenditures on civil conflict onset for the same time period. We use the Total public spending variable, the most comprehensive measure of public expenditure, which allows us to explore whether higher levels of public spending in general are less conducive to the emergence of civil conflicts. It is a continuous measure of total public expenditure as a percentage of GDP that includes basic social spending expenditures as well as defense and capital investments. This variable is obtained from the World Bank (2009). According to Models 1 and 3 in Table III, general government consumption is not significantly associated with lower likelihood of civil conflicts. This finding offers indirect support for our argument that only certain types of public spending (i.e. welfare spending) might undermine civil wars. The governments that allocate a larger share of their budget to social welfare policies with direct social insurance and assistance roles enjoy higher support from citizens and, in turn, face a lower risk of dissent. The second public spending variable examined in Table III is public expenditures on the military. Some scholars perceive military spending as an unproductive use of scarce economic resources (Henderson & Singer, 2000; Gyimah-Brempong, 1992). They suggest that the level of military spending should be positively associated with domestic unrest for two reasons. First of all, military spending has a detrimental effect on economic development, growth, investment, and factor productivity (Collier & Hoeffler, 2004; Hegre & Sambanis, 2006). Scholars also suggest that there is a trade-off between military expenditure and civilian investments, and military cuts tend to be resilient to budget cuts. As states prioritize military spending in budgeting decisions and overcompensate for security risks, they divert valuable resources away from social programs that are critical for citizens. This, in turn, contributes to poor social well-being among citizens, especially in less developed countries (Gyimah-Brempong, 1992; Adeola, 1996; Henderson & Singer, 2000). To evaluate the impact of military spending, we use the natural log of Military spending as a percentage of GDP. The data for this variable are from the Correlates of War (COW) project. According to Models 3 and 4 in Table III, we find no statistically significant relationship between military expenditure and the emergence of internal conflicts, but our Welfare spending variable remains statistically significant. This result highlights the importance of budgeting decisions in sustaining peace within a society and emphasizes the importance of our findings further. Conclusion Civil conflicts are, by and large, violent and destructive domestic events with significant transnational consequences. Unrest in one country can pose threats to the stability of neighboring countries, which undermines regional and international security. Problems posed by domestic conflicts can also lead to engagement and intervention of other states and non-state international actors. Due to their significant domestic and international implications, civil conflicts have attracted a great deal of attention from researchers and policymakers,