Group Grievances & Civil War: Some Theory and Empirics on Competing Mechanisms, 1990-2017

Recent scholarship forcefully claims that group grievances due to political exclusion and discrimination drive civil wars. This perspective argues that socio-psychological factors allow groups to overcome collective action problems. We argue that the grievance perspective (over)focuses on the ends and not means , which are critical to explain how groups survive state repression, allowing contention to become civil wars. We suggest that inclusive economic governance reduces investment in state-evading infrastructures for quotidian economic reasons. Our analyses show that group-grievance-generating political factors are poorer predictors of civil war compared with economic freedoms measured as free-market friendly policies and the private ownership of economies. These results are robust to several alternative models, data


INTRODUCTION
Much is written on the political exclusion of individuals and groups-ethnic, class, caste, gender, the ruralurban divide-and political and economic failure (Acemoglu and Robinson 2012, Alesina, Easterly and Matuszeski 2011, Bates 2001, Easterly 2006a, North, Wallis and Weingast 2013, Sen 2006).Recent arguments connect exclusionary processes in politics directly to the risk of serious armed violence and civil war, focusing specifically on group identity (Cederman, Wimmer and Min 2010, Cederman, Gledistch and Buhaug 2013, Wimmer, Cederman and Min 2009a).These studies see conflict driven by grievances rather than due to rational choice arguments based on the feasibleness for organizing costly violence (Anderton and Carter 2019, Blattman and Miguel 2010, Collier et al. 2003, Collier, Hoeffler and Rohner 2009).Indeed, strong policy statements by leading global institutions now reflect the new consensus-political inclusion and the reduction of group grievances are the paths to peace (United Nations and World Bank 2018).We reexamine this debate using the latest available data generated by theVarieties of Democracy (VDEM) project that measure group exclusion along various relevant dimensions, but we argue that what really matters are the opportunity factors driving the logistics of armed violence located in policies that constrain economic freedoms.We argue that the real mechanism explaining violence lie in economic rather than political exclusion, and the link to costly armed violence is determined by investment in state-evading infrastructures rather than in societal grievances determined by political exclusion.In other words, the origins of civil war are in exclusionary processes in economic governance that generate shadow organizations that form rebellion-specific capital rather than mobilization based on justice-seeking grounds, such as political emancipation.
Using data available for the post-Cold War decades, which in many respects should be the most reliable data generated on the basis of expert judgement, we explicitly address the question of whether political grievances matter more than economic opportunities for sustaining lengthy enough fights with states, which then enter our data sets as civil wars.Our analyses show that group-grievance-generating factors such as political exclusion and discrimination are poor predictors of civil war compared with economic freedom, our preferred measurement of economic exclusion and repression.These results are robust to several alternative models, data, and estimating method.It is hard to see why economic freedoms reduce grievances more than do political ones, if we accept that grievances are what drive civil war.Contrarily, there is good evidence to suggest that economic repression drives organization in the shadows that become the means of violence, regardless of motive.Next, we critically examine some mechanisms explicit in extant theory, offer a counter explanation, and then subject standard data to some rigorous analyses.

THEORY
Pinning down mechanisms explaining civil war is critical for getting policy right for ending armed conflicts and preventing its recurrence.The theoretical literature is split along a "battle line" that is non-trivial in terms of formulating targeted policy for preventing civil war.Since organizing violence, especially against states, is very costly, the question of "what" exactly motivates it and "how" motives translate into action (means) must be keenly understood.Despite the theoretical and empirical uncertainties, a recent broad survey of the literature authoritatively claims that "inequality, exclusion and a sense of injustice" drive armed conflict (United Nations and World Bank 2018).The recent grievance perspective, thus, is not just academic, but it now informs policy in important ways.While the report does pay some lip service to "looting and shooting" and criminalized violence, it focuses mostly on the "ends" for why people kill each other rather than the means through which this costly and complicated phenomenon happens.
The means of organizing violence against superior forces of states is the "binding" constraint-plunder and criminality, after all, exists in all societies to some extent, presumably policed by regular authorities.Sustained large conflicts that generate high enough battle deaths to be coded as civil wars in our datasets, however, require some explanation.Thus far, the grievance perspective assumes that a discriminated group simply generates enough motive andmeans for overcoming the collective action problem.The best explicated theory, however, still remains thin on the "ends/means" problem.First, we address our concerns about the recent perspectives that argue that political exclusion and discrimination underly civil war.Thereafter, we lay out an organizational logic for violence based in economic factors and policies and outline our empirical strategy for addressing the contrasting arguments.
The economics of civil war literature models the causes of political violence within a rational choice framework, where individuals make decisions based on the costs and benefits of organizing armed violence (Anderton and Carter 2019).Long drawn out contests with states require groups to have access to resources (finance) whatever the "cause" that may motivate violent organization because continued opposition to a state requires one to survive.The Collier-Hoeffler model of civil war internalizes the ends and means problem in a simple theoretical model.Rebellion is likely to occur where there are lootable resources because groups can "earn" from access to loot and fund their survival against other claimants (states, other rebels).Indeed, many of the post-Cold War conflicts, especially in Sub Saharan Africa, seemed to be driven by "loot-seeking" rather than "justice-seeking" motives, even if these conflicts had antecedents rooted in Cold War era political struggles.In any case, conflict had to be for private gain rather than for the provision of a public good.Even overtly ethnic conflicts in places such as the former Yugoslavia could be thought of as conflicts for private gain by well positioned individuals and groups who stood to profit from organizing violence and where ethnicity is instrumentalized (Gagnon 2004, Mueller 2000).
The Collier-Hoeffler model and the World Bank studies focused on "feasibility," or "opportunity" factors as being decisive for explaining civil war were criticized on empirical and theoretical grounds (Cederman and Girardin 2007, Cramer 2002, Fearon 2005).First, the empirical findings associated with natural resources and civil war were criticized for capturing aspects of natural resource dependence and the political problems associated with the resource curse ( de Soysa 2002, Fearon 2005).Fearon (2005) argued that natural resource wealth simply captured the effects of "political Dutch disease" and the link to armed conflict wasn't because of rebels solving their "ends-means" problem but because state institutions and organization happened to be weak under conditions of resource wealth.Thus, any group with anti-state feelings, or for other reasons, often along ethnic and territorial lines, could easily mobilize against a weak state.Indeed, these studies showed that it was the production of petroleum that mostly explained the resource curse, not simply because of looting rebels but also because of the administrative capacities of states under resource-wealthy conditions ( de Soysa and Neumayer 2005, Fearon and Laitin 2003, Ross 2006).Regardless of the underlying mechanisms, these arguments broadened the idea of "feasibility" (opportunity) as the underlying explanation for why some countries experienced organized armed violence and others don't.State capacities and rebel capacities must be addressed from a dyadic perspective.
Arguments about the feasibility of conflict generally focus on factors that allow rebellion against a state (means).Arguments about grievances as motive for civil war focus on "justice" as the "end" for which people mobilize armed violence against a state.These arguments focus far more on the "ends" dimensions.Further, they focus much on group identities, arguing that ethnic and other nationalisms drive violence because of political failure.Apparently, conflict occurs because some ethnic groups might be excluded from political power, driving these groups to mobilize against groups in power (Buhaug, Cederman and Gleditsch 2014, Cederman, Wimmer and Min 2010, Wimmer, Cederman and Min 2009b).Armed conflict, thus, is organized against a state due to motives driven by horizontal inequalities, which breed grievances and the search for "justice."These studies go beyond previous studies that modelled ethnic enmity based on the probabilistic measure of ethnic fractionalization by pushing theoretical arguments of how ethnic groups might organize violence against a state.These perspectives too, however, generally take the "means" as given, mostly explaining the motives.They assume that states simply cannot repress grievances enough to prevent civil war.Just causes, in other words, generate violence simply because of the effectiveness of mobilization by other-regarding, altruistic ethnic leaderships.
The grievance perspective abandons the rational choice framework for a more socio-psychological perspective.Rather than individual decisions on whether to join a rebellion based on rational criteria such as earnings potential from loot, the socio-psychological position rests on subjective factors, such as questions of allegiance to a leadership, an affinity for cause, group feelings-whether justified or not, constructed or not.The grievance perspective might suggest that the broader the grievances the greater the chance that rebellion becomes "cheaper" because of access to recruits.Indeed, conflict research since at least the 1950s has seen much debate about the "deprived-actor" model of conflict, which seemed to explain the communist-inspired rebellions throughout much of the post-War years.The primary focus of these studies was on income inequality.Apparently, inequality raised the risk of civil war because those who "have" exploit the "have nots," leading to revolt and revolution.Despite years of speculation, however, thorough empirical analyses of civil war and violence find little evidence linking income inequality to the risk of societal violence (Collier, Hoeffler and Rohner 2009, Fearon and Laitin 2003, Lichbach 1989, Neumayer 2005, Weede 1998).These studies conclude that, given the findings, the typical grievance-based propositions about armed conflict and interpersonal violence might be mistaken.As some others argued, income inequality is also associated with rising incomes in general, and higher levels of inequality possibly signal to "have nots" that their turn is next, pacifying social relation-the so-called "tunnel effect" (Hirschman and Rothschild 1973).
Relative deprivation arguments about grievances and violence are considered to be too fuzzy because grievances are hard to measure by reasonable objective standards, hard to justify from rational choice perspectives, and often difficult to distinguish as a cause rather than a symptom of violence (Lichbach 1990, Theuerkauf 2010).Moreover, empirical studies that have specifically addressed religious repression as a proxy for grievances find little support for the view that they cause religious conflict (Basedau et al. 2015).Further, studies of recruitment during ethnic civil wars do not find ethnic enmity or nationalistic feeling to have been a strong motivating factor for joining the fighting relative to much narrower solidarities, at least in the case of the Irish Republican Army and the Bosnian-Serb army (Malesevic and Ó Dochartaigh 2018).What motivated people were not high nationalist appeals, but rather everyday friendships and associations that bound people to violent actions.Importantly, the ability to organise successful violence could very easily determine a causerather than the other way around, nor is it an easy task of determining the true nature of the depth and breadth of support for violence over other solutions, which could indeed be determined by how people view their chances of success.Indeed, peaceful ethnic and other group accommodation occurs routinely in most societies (Laitin 2008, Sen 2006).
The conceptual objections to the new grievance perspectives notwithstanding, we also object to some of the ways in which these studies have been operationalized and key variables measured.First, some have taken the position of modelling grievance-based violence at the group level, by taking the ethnic group that has been able to rebel as their unit of analysis (Cederman, Weidmann and Gleditsch 2011).In other words, those groups that are deemed "relevant" are defined by their ability to challenge a state, not necessarily by some objective factor traceable to a grievance as such.Feasibility of rebelling is then traced back to factors, such as income inequality or exclusion from politics, or other such exclusionary process.Since rich people living in castles rarely if ever rebel against other rich people living in castles, almost by definition, a group that organizes violence against a state, ethnically distinct or not, must be poorer, much in the same way that any rebellion that starts off in the countryside (for strategic reasons) must necessarily be a "peasant" uprising.Clearly, this is not the case.Why such movements rarely occur despite such great disparities in wealth and access to state power, not to mention why only some ethnic groups are able to rebel while some others suffer in silence, are issues that need careful unpacking.Picking group A because it makes a claim against a state might say more about "means" than "motives."Our primary objections, thus, to these illuminating studies, thus are conceptual in terms of the mechanisms involved, as well as empirical in terms of how key variables are operationalized.The use of the "Ethnic Power Relations" data, where ethnic groups are coded on the basis of a claim, remains problematic on many counts (Marquardt and Herrera 2015).
Our study takes a critical view of the opportunity perspective s operationalization of the ends/means problem as well as the grievance perspective s measure of ethnic exclusion.First, we think that limiting motive and opportunity to the availability of natural resources is far too restrictive.We think that most poor countries governance structures invite predatory behavior and the accumulation of conflict specific capital on which violence ultimately rests-whatever the motive.Our micro-perspective locates the true cause of why some countries suffer armed conflict on questions related to economic governance that allows certain logistical bases of violence to form.Poor countries typically have predatory governments (rulers), whether democracies or autocracies, and economic power often determines who occupies offices of government.No man rules alone, therefore, most poor countries have "elites" that decide on policies and institutions that perpetuate the economic and political power (and bonds) within these elite circles.Some of these elites were created during colonial times, others took office (mostly by force) in post-colonial periods.The policymaking for narrowlydefined groups perpetuate the vicious cycles of poverty and extraction (Acemoglu and Robinson 2012, Bueno de Mesquita and Smith 2011, North, Wallis and Weingast 2013).These environments naturally contain high levels of grievance and dissent, but also high levels of deference to political elites-opposing whom can be costly.How dissent translates into violence depend on other factors.
We outline why macro-level economic governance generate the conditions that spawn violence-not because of subjectively-or objectively-derived grievances but because of how organizational bases of the logistics required for large scale armed violence form.These organizational bases exist due to quotidian activity of ordinary people outside of regulated economic life-shadow-economic activity.These largely-apolitical organizations can be often violent within (criminal) and peaceful outside, but they can under some circumstances also be mobilized to challenge the big guns of the state and survive long enough to become civil wars and insurgency.These groups might in many cases be ethno-territorially based or heterogenous in terms of identity groups.Or they might aspire to political power and be ideologically motivated, but they share the same logistical bases (command and control apparatus) as do apolitical groups.In other words, these groups exist, not so much because of the "ends" they may share, such as the search for justice, but because of the "means" they possess.Quite simply, such groups exist because they survive violent encounters with other groups, including state militaries.More often than not, groups that challenged states may have vanished before they enter our datasets.Poor country prisons are filled with "rebels" without causes.Indeed, every Mexican drug lord with a small army could potentially be an "ethnic martyr" if indeed they could claim a distinct ethnicity.
Consider the simple empirical fact that oil wealth generates violence.Oil is lootable on many conceptual levels-one can control government and loot the rents from production-or one might easily raise enough finances to capture oil wells and oust a sitting government.Thus, access to finance should be high enough to generate large deaths given the importance of the stakes.Yet, very poor countries with nothing to grab also fall into conflict.Moreover, oil wealthy countries do keep the peace for long periods of time, often with powerful external backing.Thus, anyone planning on taking over an oil wealthy country also needs the logistics for surviving, presumably by controlling an army, and fighting off powerful backers.Osama Bin Laden could have easily taken over Saudi Arabia, but then he would have had to fight the United States, something far more easily achieved in Afghanistan rather than in Saudi Arabia-in other words, there is an organizational logic for survival rather than only a simple end/means explanation.
Our argument is that poor economic policies in poor countries generate lootable income in shadow economy environments.These shadow economy environments not only deal with "illegal" goods, such as drugs and arms, but more often than not, they exist because of large payoffs accruing from state-evading activity, such as smuggling of consumer goods.These everyday activities are embedded in communities, but they also form communities.Naturally, a world that is created by bad policies must also drive societal antipathies towards rulers and vice versa.When open armed violence begins, however, it is these pre-existing logistical bases that prolong and intensify armed conflict.
Identifying and measuring logistical bases of armed violence based in shadow economic activity is no easy task.Consider that drug smuggling is associated with the payoffs for forming armed groups.However, the cocaine manufactured in Colombia causes violence throughout the region because of smuggling.Indeed, the payoffs increase exponentially the further away the drugs move from the production site so that Mexican smugglers that get the drugs to the US possibly earn the highest premium, even if the coca doesn t originate there.Similarly, identifying the intensity of shadow economic activity can only be proxied by market-distorting policies.Our empirical strategy, thus, is to compare market-distorting policy measured as economic repression (lack of economic rights and institutions) against alternative variables measuring political and economic discrimination.The relative empirical fit of the data allow us to infer the effects of alternative explanations since there is no clear logic as to why economic discrimination should be less "grievance-promoting" than political discrimination, given the intertwined nature of these phenomena.Further, to assess group grievances due to discrimination, we assess the conditional effects between state control over an economy and grouplevel discrimination along both access to jobs and access to state business opportunities.If people are denied state jobs when the economy is in the hands of the state, then we expect grievance based conflict to be most likely.We propose to use standard models of civil war and data typically used in similar studies for comparison with previous work.We test the following basic hypothesis: H1: Economic repression increases the risk of civil war relative to group-based political discrimination and exclusionH2: The effect of equitable access to state jobs and business opportunities on the risk of civil war is conditional on the extent of state ownership of the economy

DATA & METHOD
Our dependent variable is the onset of civil war from the Uppsala Conflict Data Program (UCDP) (Pettersson, Högbladh and Öberg 2019).An onset is coded when at least 25 battle-related deaths have occurred in a conflict between a rebel(s) group and a state, which may or may not involve a foreign third party (internationalized civil war) (Gleditsch et al. 2002).We purposely limit our analyses to the post-Cold War years since our main independent variable and the dependent variable are affected directly by Cold War geostrategic competition of the superpowers, not to mention the implications of this rivalry for the logistics of violence.Additionally, the most recent years (roughly 3 decades) provides the most reliable expert coding of data and the widest coverage in terms of truly politically independent countries.Nonetheless, we directly model the history of conflict stretching back to 1946 or the time since independence of former colonies.11Conflicthistory is typically modelled by entering a count of the number of years of peace since the last civil war (or since independence, or the year 1946) and a set of natural cubic splines that smooths the baseline hazard stretching back over time.Beck, Nathaniel, Jonathan N. Katz and Richard Tucker. 1998."Taking Time Seriously: Time-Series-Cross-Section Analysis with a Binary Dependent Variable."AmericanJournal of Political Science 42(4):1260-88..We use a 2-year hiatus in the fighting for coding a new onset, although our results are robust to using onsets based on the 5-year hiatus.
Our main independent variables are taken from the VDEM dataset and they measure various ways in which groups might be excluded from political power and state goods, such as state jobs and business opportunities.These exclusion variables are sorted on the basis of economic group (class), rural-urban divide (geography), and social group (ethnicity, religion, caste etc).According to the VDEM coders, political exclusion of social groups is assessed as: Exclusion is when individuals are denied access to services or participation in governed spaces (spaces that are part of the public space and the government should regulate, while excluding private spaces and organizations except when exclusion in those private spheres is linked to exclusion in the public sphere) based on their identity or belonging to a particular group.The point estimates for this index have been reversed such that the directionality is opposite to the input variables.That is, lower scores indicate a normatively better situation (e.g. more democratic) and higher scores a normatively worse situation (e.g. less democratic).Note that this directionality is opposite of that of other V-Dem indices, which generally run from normatively worse to better (VDEM 2021: 210).
The final index of exclusion is an aggregation of exclusion based on access to rights, access to political resources, access to public good, and access to state jobs and business opportunities.We test the index and several of its relevant component parts.Naturally, in these models, we also enter VDEM s basic democracy measure (polyarchy), which is simply electoral democracy (free, fair, and competitive).Since measures of exclusion enter VDEM s other definitions of democracy, we simply use the minimum definition.
Our second set of independent variables measures the degree to which a state s economy is free from government control and regulation.In other words, how free is the economy in terms of being influenced by markets (private enterprise and free exchange) rather than by states.First, we use the widely used Economic Freedom Index that measures a "capitalist" economy that encourages entrepreneurial activity with minimal involvement of the state (Berggren 2003, Easterly 2006b, Gwartney, Lawson and Hall 2011).This measure captures aspects of economic liberty within 5 fundamental areas using both objective criteria, such as tax and tariff rates, trade flows etc., as well subjectively derived indicators, such as the strength of property rights protection and legal institutions.The five areas measure the extent of state involvement in the economy, legal security and property rights, the independence of the central bank (access to sound money), freedom to trade with foreigners (international trade), and the extent of state regulation of private business (regulation of labor markets etc).Alternatively, we use the subjectively derived (expert coded) measure of "state ownership of the economy", which VDEM coders explain as: the degree to which the state owns and controls capital (including land) in the industrial, agricultural, and service sectors.It does not measure the extent of government revenue and expenditure as a share of total output; indeed, it is quite common for states with expansive fiscal policies to exercise little direct control (and virtually no ownership) over the economy (VDEM 2021: 185).22Thestate ownership indicator captures the extent of state ownership at lower values while higher values capture private ownership.We have relabeled this measure in the tables.
These variables capture very different things, but they also measure the underlying logic of our argument.Note also that a state that is able to own much of the economy (control, capital, land and labor) should also possess quite a bit of state capacity from the traditional sense of being able to control people through an elaborate administrative apparatus.
Empirical models of civil war risk have generally been poor at predictions, perhaps due to the difficulty of getting the exact timing of full scale conflict right.Yet, a number of variables have proven to be robustly related to the onset of civil war.First, per capita income matters strongly.Income captures both poverty-related questions; i.e. the productivity of labor etc., as well as state capacity and counterinsurgency infrastructure (Fearon andLaitin 2003, Vestby, Buhaug andvon Uexkull 2021).33Theper capita income is taken from the World Development Indicators online database.We use the GDP per capita in constant 2010$.This variable is logged to reduce the influence of extreme values.Secondly, population size, a fairly under theorized variable, also seems to matter robustly (Hegre and Sambanis 2006).Population size matters possibly because it captures geographic size and factors favoring insurgency, or it matters because of questions of state capacity and governance, or perhaps because large countries are strategically more likely to be in geopolitical diplomatic struggles and regional rivalries conducive to externally-supported anti-state movements.Regardless, we control for total population (logged).Additionally, we enter a term for oil export dependence, which is thought to influence the quality of governance and encourage insurgency (Fearon and Laitin 2003).We use the World Development Indicators online database s oil rents per GDP and generate a discrete variable taking the value 1 if oil rents to GDP are above 33% and 0 if not.Finally, following many others, we model the history of civil war by entering a term for years since the last civil war (since independence), or the year 1946, plus 3 natural cubic splines to smoothen the baseline hazard over time.We use logit analyses with standard errors clustered on countries.By clustering the standard errors, we assume that our observations are independent across units but dependent within units.Such standard errors provide correct coverage and are conservative (Angrist and Pischke 2009).In table 1, we test our basic model for the entire period stretching from 1960 to 2017 and from 1990-2017 using just the baseline model.We estimate a global sample and a sample of only developing countries for comparison.44Developingcountries include all countries that are not industrialized democracies, or what some people term the WENAO countries.WENAO stands for Western Europe (excluding former Eastern Bloc), North America (US, Canada), Oceania (Australia, New Zealand), and Japan.See Milanovic, Branko. 2016.Global Inequality: A New Approach for the Age of Globalization .Cambridge, MA: Belknap.As seen there, our results are in line with most previous studies (Blattman andMiguel 2010, Hegre andSambanis 2006).Per capita income reduces the risk of an onset, and population size and oil production increase it.Interestingly, electoral democracy is not statistically significant.These results are consistent across the two sample sizes and across the two time periods tested.Thus, our results on these variables are not influenced by Cold War and post-Cold War politics.Transforming the logic coefficients to the odds ratio suggests that a unit increase in income per capita reduces the risk of an onset by roughly 22% of the average risk of an onset.A similar increase in population size increases the risk by 54% of the average risk (or 2 x the impact of income per capita).Moving from a non-oil producer (with rents from oil smaller than 33% of GDP) to oil producer (rents larger than 33% of GDP) increases the risk by almost a multiple of 2 (190%  1 (column 3) for each of our group exclusion measures and select sub-components.The control variables are not shown due to space consideration, but they remain unchanged in comparison with Table 1.As seen there, neither exclusion by social group, by class, nor the subcomponents of social group civil liberties, representation of disadvantaged social groups or social group equity in terms of access to state jobs, and state businesses matter.These results are essentially unchanged when we test the longer time period.We cannot but reject the view that politically discriminated and excluded groups rebel based on the VDEM measurements of political exclusion, which do not require groups to make a claim to be counted (Marquandt and Herrera 2015).The control variables remain essentially the same as reported above.We turn next to our measures of economic exclusion measured as economic freedom and private ownership of the economy.**********TABLE 3 ABOUT HERE********** Table 3 presents results beginning with the VDEM measure of state ownership (larger values denote weak state ownership).As seen in columns 1 and 2, greater private ownership reduces the risk of an onset of civil war, results that are statistically significant and strongest when testing the developing countries.A unit increase in private ownership reduces the average risk of an onset by 22%.These results suggest that a state s control over an economy does not increase its capacity to control violence directed against it.In columns 3 and 4, the economic freedom index also unambiguously predicts a lower risk of conflict onset.A unit increase in economic freedom reduces the average risk of a civil war onset by 32%.Clearly, economic governance seems to significantly predict civil war compared with factors associated with political exclusion.Notice also that the extremely strong effect of country size estimated in the simple model in Table 1, is now much lower and only slightly larger than the effect of economic freedom (32 vs 36%).Next, to push the argument about organizational logics rather than grievances, we examine if the effect of economic freedom might be working through the quality of institutions rather than the mechanisms we argue.In Table 4, we enter regime corruption, government corruption and respect for property rights taken from the VDEM to capture institutional quality.As seen there, none of these measures are statistically significant, and economic freedom remains robustly significant and shows consistent effects.Notice that per capita income is always statistically not different from zero in the presence of economic freedom, suggesting again that the often-argued mechanism of state capacity due to taxable income and bureaucratic efficiency is probably less of a factor than the mechanism we highlight.Grievance proxies do not seem to matter.
To push the grievance mechanism further, we next interact state ownership of the economy, or VDEM s state ownership variable inverted, so as to capture greater state ownership, with the inverted equality of access to state jobs for social groups, which now captures discrimination in terms of access to state jobs.We do the same with equality of access to state business opportunities for social groups so that this indicator now measures the degree of discrimination against social groups for access to stat business opportunities.Table 5 (columns 1 & 2) reports the conditional effects of both interactions respectively.As seen in column 1, there is no statistically significant effect of the interaction between state ownership of the economy and discrimination on civil war onset.Indeed, state ownership shows a positive effect (consistent with results in Table 3) when discrimination is zero.The margins plots in Figure 1 shows the result graphically.In column 2, however, there is a small statistically significant effect when state ownership is interacted with access to state business opportunity.This result is statistically significant when assessed graphically, especially after state ownership above roughly the midpoint of state ownership (see Figure 2).Why access to state jobs might not generate grievances strong enough to challenge states compared with business opportunities is a mystery if the relative deprivation theoretical story is true.Contrarily, if elites of social groupings are overtly discriminated, then the chances that rebellion-specific capital builds up seems the more logical explanation.

ROBUSTNESS
In order to assess the robustness of our basic findings we do several tests using the developing country sample in the post-Cold War period.First, we assess whether or not the effect of economic governance (free-market economies) are essentially capturing the liberal nature of politics.We switch electoral democracy in our model to VDEM s "liberal democracy," which not only incorporates electoral democracy as a minimum but additionally includes many other rights-based indicators.Adding liberal democracy has no appreciable effect on the statistically significant effect of economic freedom.Liberal democracy too has a negative sign but is not statistically significant.Next, we add VDEM s egalitarian democracy, which incorporates indicators measuring both rights and access to political resources.This factor too is statistically not different from zero, while the negative effect of economic freedom remains statistically highly significant (p<.0001).Next, we check a direct measure of the strength of a state bureaucracy in terms of public administration by including VDEM s measurement of "rigorous and impartial public administration."An effective and impartial bureaucracy might not just reduce societal grievances regardless of regime type, but it may also capture state capacity effects for countering potential rebellion.This variable too is negative but statistically not different from zero, and its inclusion makes no difference to the significant effect of economic freedom.
Next, we assess the argument that there is a "democratic civil peace" relative to a "capitalist civil peace."Some argue that democracy reduces the risk of civil war if and only if horizontal and vertical constraints against the executive, such as well-functioning legislatures and judiciaries (checks and balances), are strong (Fjelde, Knutsen and Nygård 2021).The argument made is that when checks against the executive are strong, then leaders can make "credible commitments" to rebels, allowing peace.Such notions of rebellion assume that it is executives that make war and not legislatures.In other words, legislatures made up of the peoples' representatives (plus neutral judges) allow executives to commit and rebels to trust.Of course, from recent events in Colombia, where the president makes the peace, which is then rejected by the parliament and a popular referendum, or the many peace initiatives that were embarked on by successive executives in Sri Lanka, that were rejected by well-placed groups in the legislature, problematizes the theory.Consider also that dictators, that presumably have no formal constraints on them, keep the peace.Is this because potential rebels trust dictators?Regardless, using the same indicators from VDEM, we test their version of "horizontal constraints" in our model for the sample of developing countries in the post-Cold War period.55Horizontalconstraints are the average score of VDEM s indicators v2x jucon and v2xlg legcon respectively.Neither horizontal constraints, nor an interacted term with horizontal constraints and liberal democracy come close to statistical significance, nor do they displace the statistically significant negative effect of economic freedom.66Admittedly,our simple test thus far is neither a comprehensive assessment of their interesting theoretical arguments, nor a worthy replication of their empirical analyses.
Further, we conduct two methodologically oriented robustness checks.First, following recent studies, we estimate our models using the more flexible Generalized Additive Models (GAM), which allows the modelling of complex interactions and nonlinearities in the estimated relationships (Fjelde, Knutsen andNygård 2021, Vestby, Buhaug andvon Uexkull 2021).Figures 3 and 4 reveal that the relationships between our two economic freedom variables (private ownership of the economy and economic freedom) and the probability of civil war is reasonably well captured by a linear specification.The yellow area of the map denotes high estimated probabilities of civil war and the low probabilities (red) across combinations of economic freedom and electoral democracy.As evident, the estimated probability is high for low levels of economic freedom, irrespective of the electoral democracy score.For low levels of economic freedom, the estimated probability of civil war is relatively high across all values of polyarchy, while for high economic freedom it is low (red) across all values of polyarchy.This result provides further evidence that economic freedom is more important relative to any constraints that free and fair elections with universal suffrage might impose on the risk of civil war.
Finally, we leverage our panel data to study the within-country relationship between economic freedom and conflict.It is generally believed that estimates based on within-unit variation provide stronger evidence of a causal relationship, since these estimates by design account for all time-invariant confounders.However, although we have panel data, within-country estimates are hard to identify in our case because i) we have limited over time variation in our key variables, and ii) fixed effects estimates are well-known to be problematic in logit/probit models.88Thepopular linear probability model is not an option in our case since the dependent variable is heavily skewed and thus not reliable.Nonetheless, we re-estimate our models using a version of the so-called Mundlak transformation to estimate "hybrid models" to tease apart the betweenand within-relationships (Bell and Jones 2015).99For each variable we construct two new variables, one is the unit mean over the period, the other is each observations deviation from the unit mean.In the regressions, the former estimates the between-unit relationships, the latter estimates the within-unit relationships.Using this approach, we find that the within-unit relationship between economic freedom and conflict is negative, but in contrast with the negative between-unit relationship, the coefficient is not statistically significant.Given the limited over time variation, the imprecision in the within-unit estimate is not too surprising, but it is nevertheless reassuring that both the within-and between estimates are negative.The evidence taken together suggests strongly that economic freedoms trump various measures of state capacity, measures of group-level political exclusion and discrimination, and several different indicators of democratic political constraints on states.

CONCLUSIONS
Stein Rokkan, one of Norway s most famous political scientists, commented that even in Norway, considered one of the most egalitarian democracies in the world, "votes count, but resources decide."Similarly, social grievances surely count for generating any charged environment, but large-scale costly violence that survives fairly longer periods of time require massive investment of resources.The act of rebellion cannot be forged and maintained merely by opinion.This paper addresses the question of whether political exclusion and group-grievance-generating factors explain the onset of civil war better than the alternative, which is that factors generating opportunity for surviving the repressive capacity of rulers ultimately matters, regardless of the intensity or the nature of social grievances.Indeed, any conspiracy theory potentially generates enough people willing to attack a state, witnessed recently in Washington, DC in the United States, but civil wars require such groups to survive initial engagements and in many cases even thrive (The Tamil Tigers, Taliban, Hezbollah, Boko Haram, the Houthis etc).Factors that explain such survivability cannot be based only on the ends question driven by sociopsychological factors because of the free-rider problem, but more importantly, survivability requires an elaborate infrastructure capable of sustaining a lengthy fight-the means .The extant theories, by and large, fail satisfactorily to explain this.We argue that economic governance matters because market distortions lead to investment in shadow (illegal) economic activity that form the logistical bases of long drawn out violence.The empirical evidence seems quite clear, political exclusion and discrimination seem to matter a whole lot less than does economic repression.Why one would be moregrievance causing than the other is hard to explain, but how one generates the means with which fights become feasible is far more easily fathomed.Peacemaking that ignores the reality of means over ends is likely, thus, to fail.This reality observable in the dismal history of peacemaking, post-rebellion justice, and ethnic power sharing and other such interventions (Blair, Blattman and Hartman 2017, Kreutz 2010, Ottmann 2020).
).Next, in Table2, we estimate this basic model from Table

Table 2 .
Logistic regression estimates of a variety of political exclusion and discrimination indicators on the risk of civil war onset, 1990-2017.Dependent variable = civil war onset Political exclusion by social group Political exclusion by socio-economic group Inclusion of disadvantaged social groups Equality of civil liberties by social group Equality of power distribution by social group Access to state jobs by social group Access to business opportunities by social group Access to public services by social group Constant Number of countries Observations Robust standard errors in parentheses * p<0.01, ** p<0.05, * p<0.1 Controls estimated: Income per capita, population size, oil production, electoral democracy &

Table 3 .
Logistic regression estimates of the effects of economic freedom and private ownership of the economy on the risk of civilwar, 1990-2017

Table 4 .
Estimating the effect of the quality of governance relative to economic freedoms on the risk of civil waronset, 1990-2017

Table 5 .
Conditional effects of state ownership of the economy and targeted discrimination of social groups on the risk of civilwar, 1990-2017