Conflict and Fragile States in Africa

Conflict and Fragile States in Africa J. Paul Dunne and Nan Tian n 274 July 2017 Working Paper Series African Development Bank Group

Working Paper N o 274 Abstract This paper considers the determinants of conflict in Africa. It revisits the greed grievance debate to consider the specific regional context and changing nature of conflict in sub-saharan Africa. This is a literature that has grown rapidly in economics and political science, but some recent developments in modeling and conceptualization are providing important new contributions. It proposes and uses modeling techniques that deal with the problem of excess zeros, revisits the definition of conflict, and improves upon some proxy measures. Understanding the nature of conflict in Africa is vital to designing post-conflict economic policies and interventions, to ensure policies can prevent conflict-affected states from returning to conflict or remaining fragile. This paper is the product of the Vice-Presidency for Economic Governance and Knowledge Management. It is part of a larger effort by the African Development Bank to promote knowledge and learning, share ideas, provide open access to its research, and make a contribution to development policy. The papers featured in the Working Paper Series (WPS) are those considered to have a bearing on the mission of AfDB, its strategic objectives of Inclusive and Green Growth, and its High-5 priority areas to Power Africa, Feed Africa, Industrialize Africa, Integrate Africa and Improve Living Conditions of Africans. The authors may be contacted at workingpaper@afdb.org. Rights and Permissions All rights reserved. The text and data in this publication may be reproduced as long as the source is cited. Reproduction for commercial purposes is forbidden. The WPS disseminates the findings of work in progress, preliminary research results, and development experience and lessons, to encourage the exchange of ideas and innovative thinking among researchers, development practitioners, policy makers, and donors. The findings, interpretations, and conclusions expressed in the Bank s WPS are entirely those of the author(s) and do not necessarily represent the view of the African Development Bank Group, its Board of Directors, or the countries they represent. Working Papers are available online at https://www.afdb.org/en/documents/publications/working-paper-series/ Produced by Macroeconomics Policy, Forecasting, and Research Department Coordinator Adeleke O. Salami Correct citation: Dunne J. P. and N. Tian (2017), Conflict and Fragile States in Africa, Working Paper Series N 274, African Development Bank, Abidjan, Côte d Ivoire.

Conflict and Fragile States in Africa 1 J. Paul Dunne and Nan Tian JEL Classification: D74; C3 Keywords: Civil war; zero-inflation; greed; grievance 1 J. Paul Dunne: School of Economics and Southern Africa Labour Development Research Unit (SALDRU), University of Cape Town, Cape Town, 7701; email: john.dunne @uct.ac.za Nan Tian: Stockholm International Peace Research Institute (SIPRI) and SALDRU, University of Cape Town; email: nan.tian.eco@gmail.com 1

1 Introduction Civil war has been commonplace over the past 60 years, but until relatively recently received little attention from economists. It is not just common, but also persistent, wars lasting longer, decade after decade (Fearon, Kasara, and Laitin 2007). Blattman and Miguel (2010) estimated that, since 1960, 20 percent of all countries have experienced at least 10 years of civil conflict. Civil conflict is devastating to the countries affected, with Collier et al. (2003) suggesting that the destructive forces of civil conflict could be large enough to explain the income gap between the poorest and richest nations. One could almost see civil war as reversing development, diverting resources from productive activities to destruction and having both devastating direct costs and opportunity costs, from the loss of productive resources (Collier et al. 2003). These potential costs make it important to understand why conflicts start, and the contribution by Collier and Hoeffler (2004), which sought to empirically test two competing theoretical hypotheses concerning the determinants of intrastate armed conflict opportunity versus grievance led to a large empirical literature. Their finding of overwhelming support in favor of the view that rebellion is motivated by greed or opportunity, led to a vigorous debate and an impressive empirical literature, that generally accepted the dominance of greed, but became rather more nuanced (Blatmann and Miguel 2010). Researchers started to accept the general framework, but to examine other potential determinants that had not already been considered. Nowhere is this move beyond greed and grievance more evident than in quantitative studies of conflict prevalence in sub-saharan Africa where the imposition of artificial state borders, living in bad neighborhoods and warmer temperatures (increasingly so, in the face of climate change) have come to take central focus as explanatory variables of interest in the econometric models employed in these studies (Hendrix and Glaser 2007; Burke et al. 2009; Greene 2010). A number of developments have led to a point where there is some value to be gained from revisiting the debate. First, there are obviously more years of data available, more economic shocks, and more conflicts. This provides both more degrees of freedom and potentially more leverage for empirical analyses. Second, there have been significant improvements in the operationalization of difficult-to-measure indicators of grievance (i.e., income inequality, ethnic divisions). Third there has been some development in the estimation methods available for the analysis. In particular, is the recognition that simple probit, or logit model, does not perform well in situations where there are a large number of zeroes in the 2

dependent variable, a likely case for civil conflict since fortunately many country-year observations are zero (e.g., peace) (Dunne and Tian 2014). Another important issue is the general consensus that, while conflict is detrimental to economic development, its impact could be even greater for fragile states (Rodrik 1999). Fragile states are weak and failing with severe handicaps in economic development and institutional governance, and a conflict can be particularly devastating to these states and their neighbors (Dunne and Tian 2014). This combination of fragility and civil conflict is a particular concern for Africa, and this paper revisits the greed grievance debate and fragile states, using a database of 33 African countries for the period 1960 to 2013. A review of the determinants of civil war literature and the greed grievance debate, together with a review and definition of fragile states, is provided in the next section. Following is a discussion and outline of the estimation procedures to be used in particular, the zeroinflated Probit (ZiOP) model. Section 4 then presents the data used and variable construction and provides some descriptive statistics, followed by the empirical estimates of a greed grievance model, using the usual methods and the ZiOP model, with various robustness checks. The final section offers some conclusions, with discussion on the policy implications. 2 Causes of Civil Conflict There are a range of theoretical perspectives that inform the analysis of civil wars, reflecting the interdisciplinary nature of the research and the relatively late involvement of economists. Political scientists focused upon the grievance determinants of conflicts, with theories emphasizing how modernization could lead to disruption of social order, with social and economic change causing the breakdown of social cohesion and alteration of perceptions. This can lead to disadvantaged groups feeling threatened, identifying with ethnic or social groups and mobilizing for change. It can also ignite of old hatreds, and in times of change and crisis, successful ethnic entrepreneurial networks, despite their importance for economic growth, can become a source of resentment. A formalization of this perspective was provided by political rational choice theories. These focused on the role of political repression, failing institutions, transitions, and informational problems, which together with a failure to redress grievances economic or political can lead to conflict. An alternative was provided by constructivist theories, which focus on the social construction of identity, rather than accepting it as some fixed attribute. It is then political mobilization that leads to civil violence, with leaders constructing ethnic and social identity in ways that benefit themselves (Sambanis 2002). 3

In contrast, the focus of economists was the greed-based determinants of conflict. Grossman (1991) modeled rebellion as an industry, where insurgents gain profits from looting and have costs. In this model, insurgents are no different to bandits, and the incidence of rebellion is fully explained by atypical circumstances that generate profitable opportunities. In a different line, Hirschleifer (1995) asked why fighting occurred between rational agents, given it was a Pareto suboptimal situation. A model with preferences, opportunities, and perceptions suggested it was possible that opportunities and grievances were wrongly perceived because of asymmetric information. Agents overestimate capability and their probability of winning, and grievances are based on limited or inaccurate information and/or divergent preferences. While economists generally assume property rights are exogenous, a limited body of theoretical work has focused on endogenous property rights and economic theories of appropriation. Rational agents exist in a lawless setting with competition for resources, and predation and defense are alternatives to directly productive activities. Without well-defined property rights, contracts cannot be enforced, except by a ruler or hegemon, and they can be replaced forcibly. In the contest model, an incumbent and a rebel group allocate resources to production or appropriation through a production function, while a contest success function determines the probability of success, based on the proportion of arms and the effectiveness of technology in an environment of informational asymmetry. The problem with this model is that it always gives a fighting outcome (Skaperdas 2001; Dunne and Coulomb 2008). These economic theory perspectives suggest that the onset of civil conflict is linked to the ability of insurgents to make a profit the greed hypothesis rather than the result of grievances. Collier and Hoeffler (2004) provided an empirical analysis of the competing hypotheses, suggesting that while political grievances are universal, economic incentives are not, and so are often decisive in the start of conflict. They use a Hirschleifer-type framework (1995), where rebellion is the outcome of rational decision-making, subject to constraints of the rebel labor market. The probability of rebel victory depends on the ability of the incumbent to defend, which is determined by technology (and although the technology is also available to the rebels, it is limited) and military spending, to which the rebels do not have access. So the taxable base can be used as a measure of how strong the government is, but it also measures the potential gains that rebels can make if they achieve victory. Population is then used as a proxy for the rebels desire for secession, while per capita income and the duration of the conflict serve as proxies for the costs of rebellion, both opportunity costs and economic disruption costs. Measures of fractionalization are used as proxy for the costs of coordination, 4

and the transaction and coordination costs to a rebellion are approximated by ethnic fractionalization. Expected utility is then a function of the probability of victory, the size of the taxable base of the country, and GDP per capita. The expected gain from the rebellion is an increasing function of the size of population, and the probability of conflict will be expected to decrease with GDP per capita and expected duration. Rebel groups have the potential to finance an insurrection if they can extort natural resources, get donations from diaspora, and get support from hostile foreign governments. The ratio of primary commodity exports to GDP natural resources in exports, the proportion of the population that moved to the United States, and the indicator variable for Cold War all are viewed as proxies for these funding sources. Insurrection is considered more likely if individuals foregone income is low, which will mean the cost of recruits will be low, and this will be determined by economic growth and its impact on labor demand, income levels, and levels of schooling. It will also be encouraged if the cost of military equipment is low, and this is likely to be reflected in the time since the previous conflict, as the shorter it is, the more likely weapons will be readily available. Success is more likely if there is weak government military capability, and this will be determined by the nature of the terrain mountainous areas assist guerrillas and the distribution of the population, because if it is spread out thinly, it will be harder to police. It is also likely that low social cohesion will make it easier for the rebels, which can be approximated by measures of ethnic fractionalization. For objective grievances, measures of polarization are used namely, ethnic and religious differences. The Polity III data set and political exclusion is used to approximate political repression, with the Gini coeffient acting as a proxy for a measure of ethnic dominance and economic inequality. The theorists argue greed and misperceived grievance have important similarities. Opportunity and viability provide common conditions for both profit and nonprofit rebel organizations, but the groups are observationally equivalent, making it impossible to discern the underlying drive of the organization. Using five-year averages as data points, researchers estimate separate models for opportunity and grievance, finding opportunity is the one that seems to works best. They then combine the two models, finding that the probability of conflict is mainly determined by slow growth, and the importance of natural resources, which also increases the probability, as well as foregone income, measured by secondary school attainment, which also reduces the probability as the attainment is increased. The proxies for ethnic fractionalization, inequality and democracy, are all insignificant, leading the theorists to conclude that the greed factors 5

dominate the grievance, ones. Factors that influence opportunity, finance, cost of rebellion, and military advantage are significant in determining civil war, while most proxies of grievance are insignificant, though population has an effect and time seems to heal. The finding that opportunity explains conflict risk is supportive of the economic interpretation of rebellion as motivated by greed (Collier and Hoeffler 2004; 2007). Around the same time, Fearon and Laitin (2003) developed a slightly different model, a one-shot, reduced-form game of insurgency, where the size of a rebellion is influenced by government effort and the scale of the initial rebellion. They also found that political grievance had little explanatory power, but that state institutional capacity was significant suggesting that wars are caused by countries having weak institutions. They differed with Collier and Hoeffler (2004) in interpreting GDP per capita as reflecting state capacity, rather than as an opportunity cost. They also differed in how they coded civil wars and used annual data, rather than five-year averages. The two papers had major impacts on research and debate and led to a large literature that has advanced understanding, telling us what we do not know, as well as what we do. Clearly, it is not enough to assume grievance drive conflicts, and greed or economic factors have to be considered (Blattman and Miguel 2010). While the general cross-country consensus in the literature developed that motivations of greed outweigh those of grievance in explaining civil war onset, the literature continued to develop and improve in a number of areas. First, political scientists have questioned the apparent lack of significance of variables that are proxies for objective grievance. In a recent contribution by Buhaug, Cederman, and Gleditsch (2014), the authors argued that the lack of significance had to do with the poor proxy variables used in previous research and showed that better proxies indicate that grievances do matter. This has led to efforts being made to improve measurement and obtain better proxies, with natural resource data being improved for resources and better measurement of grievance, better measurement of inequality to include horizontal and vertical inequality, and better measures of weak institutions (Lujala, Gleditsch, and Gilmore 2005; Wucherpfennig et al. 2011). Second, some attempts have been made at improving causal identification. The potential endogeneity of GDP to conflict led to the use of rainfall as an instrument, given that it will impact on agrarian economies but not on conflict. Other attempts have used price shocks and trade shocks in a similar manner. The identification problems remains an issue, mostly due 6

to difficulties in finding appropriate instruments (Blattman and Miguel 2010; Miguel, Satyanath, and Sergenti 2004). Third, some attempts have been made to consider possible spillover effects of conflicts, creating conflicts in other countries, with the feedback of refugees keeping conflicts going (Salehyan and Gleditsch 2006; Dunne and Tian 2014). Fourth, questions have been raised about measures of conflict and violence. In the past, war tended to be determined as an event in which there were more than 1,000 battle-related deaths, with peace defined as less than this. Initially, this was developed for inter-country conflicts and continued in use, even after the focus shifted to civil conflicts in the post-cold War world. This was seen, then, as unsuitable, and a definition of conflict, of when there were more than 25 battle-related deaths, was applied. A further development saw Besley and Persson (2010, 2014) create a nonbinary ordinal measure of civil violence, with 0 as the value for peace, 1 for civil repression, based on Banks (2005), and 2 for large-scale civil conflict with more than 1,000 battle deaths. 2 Fifth, there has been some concern over the estimation methods used. These have generally used a zero one dependent variable for conflict, but this leads to a lot of zeros (peace years), and they are likely to be from different data generation processes. There is a difference in interpretation between a a country that is in and out of war and a country that is generally at peace, so a 0 value in a particular year for Botswana is rather different to that for the DRC (Bagozzi et al. 2015). In this study, the focus is, first, on analyzing the determinants of conflict onset in Africa by applying a comprehensive greed grievance model to more up-to-date data and using better proxies that have been developed over time. Second is to consider the measurement of conflict onset and use the wider definition mentioned here, together with relevant econometric methods. 3 Data To operationalize a general greed grievance empirical model, a range of variables were collected following developments in the literature. Proxies for greed or opportunity include real GDP, growth in GDP per capita, degree of urbanization, life expectancy, and natural 2 New datasets are allowing more consistent and detailed information to be used, such as the data set of global instances of political violence (http://ucdp.uu.se/ged/). 7

resource dependence. For this study, two sets of income variables were collected from the World Bank and Penn World Tables 8.0. Degree of urbanization is measured as the proportion of a country's population living in an urban environment, while life expectancy follows the usual measurement. 3 Male secondary school enrollment was not used in the estimations due to poor and incomplete data. Following from the literature, natural resource dependence is measured by the share of primary commodity exports in GDP. The World Bank provides data for the period 1960 to 1999, which was cross-referenced with Fearon (2005) for consistency. The remaining 14 years are constructed, using export data (primary commodities) provided by the World Trade Organization (WTO), and GDP, from the World Bank. Given the ongoing debates on the measure of natural resource dependence and the type of commodities used, three additional measurements are considered. A measure of oil production in metric tons and oil exports greater than one-third of total exports are used as proxy for oil abundance and dependence, respectively. 4 To distinguish fuel and nonfuel minerals from other primary commodities, a mineral dependence variable was created. A country is considered mineral dependent, if its mineral exports constitute 25 percent or more of a country's total tangible exports. Percentage of mountainous terrain in a given country is included, as an indicator of military accessibility or safe havens for rebels. The grievance variables are, for the most part, common to those identified by Collier and Hoeffler (2004) and Fearon and Laitin (2003). This paper considers three general measures of grievance: ethnic and religious hatred, political repression or freedom, and income inequality (horizontal inequality). Ethnic fractionalization is the most commonly chosen indicator to test the linkage between ethnicity and civil conflict. 5 Measurements of ethnic fractionalization is taken from Collier and Hoeffler (2004), with ethnic dominance, which is measured as a binary variable, taking on the value of one if the largest ethnic group in a country amounts to 45 90 percent of the population, and is used as alternative to ethnic hatred. To measure religious 3 This data is sourced from the World Bank. The degree of urbanization can also be thought of as a measurement of geographic dispersion: the greater the urbanization, the lower the geographic dispersion. All income figures are adjusted for purchasing power parity (PPP). 4 Oil production data, in metric tons, are provided by Ross (2013), for the years, 1932 to 2011. The additional two years were drawn from the same source used by Ross,, the US Department of Energy website for international energy statistics: http://www.eia.gov/cfapps/ipdbproject/iedindex3.cfm 5 Initially used by Easterly and Levine (1997), the fractionalization index follows in accordance with Herfindahl s formula, and is interpreted as the probability that two randomly selected individuals in a population belong to different ethnic groups. 8

hatred, Collier and Hoeffler (2004) constructed a fractionalization index analogous to ethnic fractionalization, and this is, in turn, used in the estimation process. Other things being equal, political democracy or freedom should be associated with less discrimination, repression, and civil war. Data from the Polity IV database is used to measure political rights, with the variable polity ranging from 10 (high autocracy) to 10 (high democracy). The relationship between political freedom and civil war has often been thought of having a nonlinear effect (Hegre et al. 2001). This hypothesis is tested through the inclusion of a polity-squared term. In a recent paper by Buhaug, Cederman, and Gleditsch (2014), the authors found that new grievance indices of horizontal income inequality and political discrimination performed much better than conventional indicators. They argued that economic grievance is captured by the relative gap between the mean national income and the income level of the poorest and richest groups (positive and negative horizontal inequality), while ethnopolitical grievance is measured by the demographic size of the largest ethnic group discriminated against. 6 This paper uses these alternative variables as substitutes in robustness checks for ethnopolitical and economic grievance. The control variables included in the model are the standard ones found in the literature (i.e., population and cold war). Finally, the dependent variable used here takes on a value of 0 for all peace year observations and a value of 1 for minor conflict years with combat deaths ranging between 25 999, and 2 for full-scale civil wars with annual battle deaths above 1,000. Table 2 presents descriptive statistics of the aforementioned variables with a breakdown by conflict experience and always zeroes or always peaceful. These results seem to support the central thesis that the different zeroes in the sample are formed through completely separate processes. For the always zero or complete peace group, GDP per capita, per capita GDP growth, rate of urbanization, life expectancy, and political freedom are all higher than the not always zero group. Moreover, countries that are potentially completely peaceful have lower levels of ethnic and religious fractionalization and income inequality. 7 Estimated correlations suggest some association between income and inequality variables and the likelihood of a country being completely peaceful versus incompletely peaceful. In episodes of civil conflict, 6 For full description and derivation of the variables, see Buhaug, Cederman, and Gleditsch (2014). 7 LDG = largest discriminated (against) ethnic group, PHI = positive horizontal inequality (relative gap between mean national income and income level of the richest group), NHI = negative horizontal inequality (relative gap between mean national income and income level of the poorest group) 9

GDP per capita, GDP growth, rate of urbanization, life expectancy, and political freedom are all lower compared to times of peace. Similarly, ethnic divisions, income inequality, and substantial amounts of rough terrain are higher in cases of civil war. Interestingly, primary commodity exports as a share of GDP is on average lower in episodes of civil war compared to no civil war. Opportunity Table 1. Descriptive Statistics Means Full Sample Always 0 Not Always 0 Civil War No Civil War Primary commodity 0.156 0.178 0.139 0.109 0.164 exports/gdp GDP per capita 7931 14069 3311 3172 8699 GDP per capita growth 0.018 0.022 0.016 0.010 0.019 Mountains 16.38 14.93 18.11 23.16 15.33 Rate of urbanization 46.94 56.00 39.73 40.61 47.92 Life expectancy 61.61 66.15 57.98 59.41 61.95 Oil production 17000 13700 19300 19100 16700 (metric tons 000 s) Mineral dependence 0.493 0.415 0.545 0.550 0.484 Oil exports 0.187 0.155 0.208 0.168 0.189 Grievance Ethnic frac (C&H) 63.02 52.06 69.85 77.47 60.05 Ethnic dominance 0.470 0.483 0.467 0.549 0.457 Religious frac (0 100) 36.47 36.07 36.58 0.36 0.37 Polity IV ( 10 to 10) 1.13 3.84-0.73 0.97 1.30 LDG 0.056 0.024 0.081 0.142 0.042 NHI 1.189 1.064 1.278 1.398 1.155 PHI 1.201 1.086 1.287 1.224 1.197 4 Greed vs. Grievance Revisited Estimating the probability of civil conflict using an ordered probit gave the results in Table 2. The civil conflict dependent variable takes the value of 1 if deaths total over 25 in a given battle, 2 if there are over 1,000 battle deaths in a given year, and 0 otherwise. 10

Table 2. Ordered Probit of Civil War Prevalence 1960 2013 (1) (2) (3) Probit Probit Probit Outcome Outcome Outcome Opportunity Pri exports/ GDP -5.329** -5.601** -4.091** (0.966) (0.978) (0.963) Pri exports/ GDP2 7.801** 9.045** 5.908** (1.585) (1.595) (1.597) log real GDP -0.024-0.137** -0.202** (0.052) (0.052) (0.050) RGDPPC growth -2.496** -2.455** -2.492** (0.523) (0.520) (0.527) log mountains 0.054* 0.118** 0.062* (0.028) (0.030) (0.028) Grievance Polity Index -0.015-0.018 0.001 (0.032) (0.032) (0.031) Polity Index2 0.001 0.003-0.005 (0.005) (0.005) (0.004) Ethno frac (F&L) 6.022** (0.998) Ethno frac2 (F&L -5.529** (0.934) Ethno frac (C&H) 0.011 (0.008) Ethno frac2 (C&H) -0.001 (0.001) Ethnic dominance 0.210* 0.292* (0.086) (0.119) Continued on next page 11

Table 2. continued (1) (2) (3) Probit Probit Probit Outcome Outcome Outcome Religious frac 0.967** 0.218 (0.301) (0.275) LDG 1.264** (0.168) PHI -01.72* (0.078) NHI 0.859** (0.125) Control Log population 0.340** 0.413** 0.514** (0.066) (0.068) (0.679) Cold War -0.024-0.008-0.043** (0.097) (0.096) (0.104) Observations 1,519 1,519 1,542 Log likelihood -941.824-944.336-901.834 AIC 1913.65 1918.67 1835.67 Notes: Dependent variable: Conflict prevalence; AIC = Akaike Information Criterion; Standard errors in parentheses; Significance levels: ** p < 0.01,* p < 0.05, t p < 0.1; LDG = largest discriminated (against) ethnic group, PHI = positive horizontal inequality (relative gap between mean national income and income level of the richest group), NHI = negative horizontal inequality (relative gap between mean national income and income level of the poorest group). The results (columns 1 3) provide the standard ordered Probit model with a number of alternative specifications considered. These additional estimates replace ethnolinguistic fractionalization used by Fearon and Laitin (2003) with that from Collier and Hoeffler (2004) (specification 2) and introduce a new ethnic discrimination measure and income inequality (specification 3). The results reveal that in the case of the normal ordered Probit, irrespective of the specification, primary commodity exports, as a share of GDP, have a nonlinear relationship on civil conflict incidence, first decreasing and then increasing. This is a finding that is opposite to that found in the existing literature, however, and in light of the summary statistics in Table 2, where primary commodity exports, as a share of GDP, are lower for 12

countries not in civil conflict, the result makes some empirical sense. GDP and per capita GDP growth are significant in decreasing the probability of civil war, while presence of mountainous terrain seems to increase civil war risk through the proxy of geographic dispersion, which inhibits government or military capacity. As for grievance variables, the results from the three specifications offer a similar conclusion: first, political freedom or opportunity does not matter the coefficient and its square term are insignificant in determining chances of civil war. Second, ethnicity matters, conditional on the choice of the ethnic fractionalization variable. The coefficient that is a proxy for ethnic grievance, which measures a systematic inequality in ethnopolitical opportunities, is positive and statistically significant, with its square term negative and significant when the Fearon and Laitin (2003) variable (1) is used, but is insignificant on the Collier and Hoeffler (2004) variable (2). As expected, ethnic dominance increases civil war risk, while religious fractionalization has a mixed effect. Given the differing results on ethnolinguistic fractionalization between specification 1 and 2, specification 3 looks for an alternative measure and also includes horizontal income inequality. The variable LGD, which is a proxy for ethnic and political inequality, is positive and highly significant. 8 Additionally, of the horizontal measures of economic inequality, the relative gap between the country-level GDP per capita and the mean per capita income for the poorest ethnic group in a given country, NHI, is positive and statistically significant at the 10 percent level, suggesting that African countries with one or more ethnic group(s) radically poorer than the national average have a higher risk of conflict onset. The opposite, however, is also true for the relative gap between the per capita GDP mean and that of the richest group, PHI, which is negative and significant. Here, income inequality has less meaning to the richest group, and there is an incentive to keep peace and retain wealth rather than to rebel. As for the control variables, population has a positive and significant effect on an African countries civil war prevalence, while the Cold War dummy is only negative and significant for the last specification. In most analyses on the determinants of civil conflict, an ordered dependent variable is used, in which a given country-year is assigned a value of 0 for peace and a value of 1 when violence between the state and another side reaches a given threshold, thereby classifying it as 8 For full explanation of the largest discriminated (against) ethnic growth (LGD), see Buhaug, Cederman, and Gleditsch (2014). 13

a civil war. This would generally mean that there are a large number of zero observations, since peaceful years will dominate conflict years. These zeros can be considered as reflecting rather different states, one where the structural and societal forces ensure a zero probability of civil conflict regardless of greed or grievance incentives and another that reflects a break in fighting and a high probability of returning to conflict. The first group of zeros will often be non-fragile states, such as South Africa or Botswana, and can be labeled complete-peace while the second group are often found in fragile regions such as Central, West, or East Africa, from which the zeroes can be labeled as incomplete-peace. The main difference between the first and second case of zero is that while the probability of transition into war for first type is zero, the probability for the incomplete peace group is not. In the case of incomplete-peace, incentives resulting from opportunity or grievance can induce violent conflict. Given the high proportion of heterogeneous zeroes in the analysis, using ordinary probit or logit models may not be appropriate tools for statistical inference and can potentially give biased estimates (Bagozzi et al. 2015). A more satisfactory estimation method is the split population or two-part model proposed by Harris and Zhao (2007) and Vance and Ritter (2014). This is typically in the form of zero-inflated models, or in this case, a zero-inflated Probit model, where estimations follow two stages. The first of the two latent equations, stage one, is a selection equation, while the second stage is a Probit outcome equation. This splits the observations into two processes, each potentially having different sets of explanatory variables. In the context of civil war prevalence, zero observations in process 0 (wi = 0) include inflated zeroes, consistent with countries that never experience civil conflict (e.g., Sweden), while zero observations in process 1 (wi = 1) includes cases for which the probability of transitioning into a civil conflict is not zero, and civil war casualties have not reached the lower bound (or limit) of 1,000 battle-related deaths. The binary variable w indicates the split between process 0 (with wi = 0 for no war) and process 1 (with wi = 1 for war). The variable w is related to the latent dependent variable w i so that wi = 1 for w i > 0 and wi = 0 for w i 0, where w i now represents the propensity to enter process 1 and is given by the split probit (1st stage) equation: w i = x i γ + μ i (1) 14

where x i is a vector of covariates, γ is its coefficients and μ i is the error term. The probability of i falling into process 1 is Pr(w i = 1 x i ) = Pr(w i > 0 x i ) = Ψ(x i g), and the probability that it is in process 0 is Pr(w i = 0 x i ) = Pr(w i 0 x i ) = 1 Ψ(x i γ), where (.) is the standard normal cumulative distribution function. For the Probit outcome equation, the propensity for participation in which the response variable Y i (i.e, conflict) has a distribution given by: w i + (1 w i ) e( λ i ), y i = 0 Pr(Y i = y i ) = { (1 w i ) e( λ i ) λ y i i, y y! i > 0 i (2) where the parameters λ i and w i depend on vectors of covariates x i and zi,, respectively, which are modeled as: and log(λ i ) = x i t β (3) log ( w i 1 w i ) = z i t γ (4) with mean and variance as E(Y i ) = (1 w i ) i and var(y i ) = μ + ( w i 1 w i ) μ 2. In this ZIP model, the matrices z and x contain different sets of experimental factor and covariate effects that relate to the probability of the zero state (zero probability of civil war) and the Poisson mean in the nonzero state (probable civil war), respectively. Thus, the s have interpretations in terms of the factor level effect on the probability that there is a zero probability of conflict and the /3 s have the interpretation of the effect on the average risk of civil war when the probability is non-zero. Following Lambert (1992), the ZIP model (Equation 2) can be regressed using maximum likelihood with an Expectation-maximum (EM) algorithm. 9 The use of ZioP model allows more accurate estimates to be obtained, compared to standard probit or logit models. The probability of a zero observation is now modeled 9 For full derivation of the model, see Lambert (1992) and Hall (2000). 15

conditionally on the probability of zero from the Probit process plus the probability of being in process 0 from the splitting equation. It should be noted that the usefulness of the model (i.e., unbiased estimates) declines when the size of the split in the sample population becomes very big or very small, leading to biased results. 10 Bagozzi et al. (2015) suggested that this becomes an issue when there is less than 10 percent or greater than 90 percent of zero observations. Estimating the general greed grievance models using the zero-inflated ordered probit model give the results in Table 3. The specification for the inflation equation is limited to GDP, per capita GDP growth, political freedom (Polity) and ethnic divisions (Eth Frac), as such factors promote a compatibility of interests between the state and its citizens, which in turn influences the probability that a country is in the always zero group and always experiences peace. That said, to ensure that the zero-inflated ordered probit (ZiOP) estimates in Table 3, specification 1, are not driven by choice of variables, a model in which all the covariates in the outcome equation are included in the inflation equation is also estimated (Table 3, specification 2). 10 Statistical inference becomes increasingly difficult as the proportional of zeroes gets close to one. 16

Table 3. Probit and ZIP Regression of Civil War Prevalence 1960 2013 (1) (2) ZiOP ZiOP Outcome Inflation Outcome Inflation Opportunity log RGDP -0.238** -0.290* -0.242* -0.480** (0.064) (0.152) (0.096) (0.147) RGDPPC growth -2.537** -1.507 0.783* -2.144t (0.550) (1.299) (0.803) (1.243) Pri exports/gdp -9.079** -12.171** 13.882** (1.122) (1.473) (2.535) Pri exports/ GDP2 11.316** 16.374** -20.059** (1.709) (2.604) (3.504) log 3 mountains 0.044 0.247** 0.781** (0.030) (0.628) (0.083) Grievance Polity Index -0.039** 0.317** -0.074** 0.336** (0.010) (0.108) (0.012) (0.110) Polity Index2-0.015** -0.028* -0.022** -0.285* (0.002) (0.014) (0.003) (0.012) Eth frac (F&L) 7.097** 2.784** 5.463** -13.739** (1.188) (0.673) (2.006) (3.048) Continued on next page 17

Table 3 - Continued from previous page (1) (2) ZiOP ZiOP Outcome Inflation Outcome Inflation Eth frac2 (F&L) -6.143** -6.934** 18.619** (1.069) ( 1.599) (3.248) Ethnic dominance 0.417** -2.106** 0.576** -0.332 (0.096) (0.467) (0.119) (0.269) Religious frac 0.488 0.734t 0.669 (0.328) (0.439) (0.786) Controls log Population 0.112** 1.818** -0.042 1.544** (0.032) (0.243) (0.104) (0.283) Cold War 0.008 0.464* (0.102) (0.212) Constant -6.312** -21.890** -2.435t -17.699** (0.977) (3.004) (1.364) (3.228) Observations 1,519 1,519 Log likelihood -859.48-820.04 AIC 1768.95 1696.08 Notes: AIC = Akaike Information Criterion; Dependent variable: Conflict prevalence; Standard errors in parentheses; Significance levels: ** p < 0.01,* p < 0.05, tp < 0.1 Estimating the probability of civil conflict, the zero-inflated ordered probit model in Table 3 can be directly compared to the ordered probit model in Table 2, specification 1. To start with, the coefficients reported in the first stage, inflation equation, of the ZiOP model reveal that GDP per capita has a negative and significant effect on the likelihood of a countryyear not being among the always-zero or peace group and then experiencing any level of civil violence. Additionally, political freedom seems to have the usual nonlinear effect of increasing the likelihood of civil conflict and then decreasing it past a certain point. Ethnicity also plays an important role in the different observed zeroes; here the more ethnically diverse a county's population is, the more likely it is to experience any form of violence, while a country with one ethnic dominant group has the opposite effect in decreasing chances of a civil conflict. 18

Turning towards the outcome equation, the estimates here are conditioned only for countries that are able to experience a civil conflict. In other words, the estimates are for country-years that are not in the always peaceful group. As mentioned previously, the key feature of this group is that there are societal forces that make the probability of transitioning into violence non zero. These would essentially be African fragile states. By distinguishing between the different types of zeroes, or types of countries, the ZiOP provides some new insights on the greed grievance estimates. While the ZiOP model (1) gives signs that are consistent with the standard ordered probit, there are substantial differences in the significance of the grievance terms. Primary commodity exports as a proportion of GDP show the same effect as before, albeit at a higher turning point of 40 percent. Income, both its level and its growth, decreases the likelihood that a country experiences civil conflict, conditional on that country being able to experience a conflict. Or in other words, a condition on it being a fragile state. Proxies for ethnopolitical grievance are better represented using the zero inflated models than the ordered probit, with political freedom now a significant predictor of civil war prevalence. This effect, however, is not of the usual inverse u-shape of first increasing and then decreasing, but rather decreases throughout. This is a very interesting finding and suggests that for the fragile states any kind of improvement in political freedom (less political oppression) will substantially lower the likelihood of any form of civil violence, with the impact diminishing as political grievance decreases. Ethnolinguistic fractionalization remains significant and of the correct sign, first increasing the risk and then decreasing when the probability of randomly selecting 2 individuals from different ethnic group reaches 57 percent. In addition to having more explanatory power and significance in the grievance variables, the zero-inflated ordered probit estimates are shown to have lower standard errors than the normal probit. An Akaike Information Criterion (AIC) test is run for both the probit and ZiOP, with the lower AIC value from the ZiOP suggesting that the zero-inflated models fits the data better than the probit. As suggested in Cameron and Trevidi (2010), all regressions are estimated using robust standard errors, while the proportion of zero observations in the sample (76.3 percent) falls within the accepted band of 10 to 90 percent (Bagozzi et al. 2015). Specification 2 represents the full model with all the covariates of the outcome equation also in the inflation equation. Here a comparison between (1) and the full model (2) is also a test for researcher s degrees of freedom, whereby a researcher could selectively choose specifications in order to generate significant results or false positives Simmons, Nelson, and 19

Simonsohn (2011). The similar results in (1) and (2) suggest the model is well specified, consistent to various specifications and not sensitive to researcher's degrees of freedom. To consider the robustness of the results, a number of alternative specifications (Table 2, specifications 2 and 3) are reestimated using the zero-inflated probit model. Table 4 in the Appendix is one example, with horizontal income inequality and ethnic discrimination added in place of ethnic dominance and religious fractionalization and Fearon and Laitin s (2003) ethnic fractionalization measure is replaced with Collier and Hoeffler s (2004). These results remain consistent with the estimates in Table 3, where the zero-inflated ordered probit model is preferred to the ordered probit model in almost all instances. Other variants of the zero-inflated ordered probit were estimated replacing primary commodity exports with either mineral dependence, oil production, or oil export; replacing the polity index with the freedom house measure, democracy, and autocracy dummies; substituting income variables with urbanization rate and life expectancy. The results are shown to be relatively robust, with primary commodity dependence increasing civil war risk, democracy political freedom, and higher urbanization decreasing civil war risk. 5 Conclusion This paper has revisited the greed-grievance debate within the context of fragility, using a data set of 33 African countries for the period 1960 to 2013. This seemed justified for a number of reasons: the existence of more years of data including more economic shocks and more conflicts, the significant improvements in the operationalization of difficult-to-measure indicators of grievance (i.e., income inequality, ethnic divisions), and the development of new estimation methods that seem well suited to the subject. It is also important to undertake such an analysis within the context of fragile states, as conflict is likely to be even more detrimental to their economic development. Thus, it is even more important to understand the drivers of conflict onset. Empirical estimations using a standard order Probit estimation technique do not account for the heterogeneous zeroes, and a zero-inflated model is also used, which distinguishes between observations that come from countries with a low probability of conflict and others. In essence, this is distinguishing between an important factor in determining fragile and nonfragile states. The main results are: first, unlike much of the earlier literature, civil war risk is not wholly dominated by greed, with grievance terms significant. Second, the zero-inflated ordered 20

probit models seem to perform better than the standard probit models and be better able to statistically account for observable and latent factors that produce different types of peace observations. The results suggest that using the ordinary probit, has biased the estimates, giving greater weighting to opportunity variables over grievance variables. This led to most empirical work finding opportunity or income variables as the main determinant of civil conflict. As one takes a deeper look at what type of country is mostly associated with the always zero or complete peace group, the answer is often high-income countries. By not distinguishing the different zeroes, the normal probit gave a likelihood of war calculation that included countries conditioned to not experience such an event. These countries main attribute is high income, and thus income variables were estimated with greater emphasis and significance, crowding out the grievance variable's explanatory power. By using a zero-inflated model and splitting the estimation process into two stages, greed and grievance variables are given equal emphasis, which makes it clear that both ethnopolitical and economic grievance matter, with substantial explanatory power in predicting civil war risk. Clearly, economic factors are important in determining conflict onset, but so are grievances, and this is clearer when the lower probability of higher income/peaceful countries is considered. In post-conflict situations, it is important to distinguish between fragile and nonfragile states and to carefully study the causes of the conflicts, both in terms of greed and grievance factors, and to deal with the underlying problems, rather than believing that general prescriptive policies will suffice (Brauer and Dunne 2012). References Bagozzi, B. E., D. W. Hill, W. H. Moore, and B. Mukherjee. 2015. Modelling Two Types of Peace: The Zero-inflated Order Probit (ZiOP) Model in Conflict Research. Journal of Conflict Resolution 59 (4): 728 52. Besley, T., and T. Persson, 2010. State Capacity, Conflict and Development. Econometrica 78(1) :1 34. Besley, T., and T. Persson. 2014. The Causes and Consequences of Development Clusters: State Capacity, Peace and Income. Annual Review of Economics 6: 927 49. Blattman, C., and E. Miguel. 2010. Civil War. Journal of Economic Literature 48 (1): 3 57. Brauer, J., and J. P. Dunne. 2012. Peace Economics: A Macroeconomic Primer for Violence- Afflicted States. Washington, DC: US Institute of Peace. Buhaug, H., L.-E. Cederman, and K. S. Gleditsch. 2014. Square Pegs in Round Holes: Inequalities, Grievances, and Civil War. International Studies Quarterly 58 (2): 418 31. 21

Burke, M. B., E. Miguel, S. Satyanath, J. A. Dykema, and D. B. Lobell. 2009. Warming Increases the Risk of Civil War in Africa. Proceedings of the National Academy of Sciences (USA) 106 (49): 20670 4. Cameron, A. C., and P. K. Trivedi. 2010. Microeconometrics Using Stata. Rev Ed. College Station, TX: Stata Press. Collier, P., and A. Hoeffler. 2004. Greed and Grievance in Civil War. Oxford Economic Papers, 56 (4): 563 95. Collier, P., and A. Hoeffler. 2007. Civil War. In Handbook of Defense Economics, Vol. 2, edited by K. Hartley and T. Sandler, 711 39. Amsterdam: North Holland. Collier, P., V. L. Elliot, H. Hegre, A. Hoeffler, M. Reynal-Querol, and N. Sambanis. 2003. Breaking the Conflict Trap: Civil War and Development Policy. Washington, DC: The World Bank. Dunne, P., and F. Coulomb. 2008. Peace, War and International Security: Economic Theories. In War, Peace and Security, Vol. 6, edited by J. Fontanel and M. Chatterji, 13 36. Bingley, UK: Emerald. Dunne, J. P. and N. Tian. 2014. Conflict Spillovers and Growth in Africa. Peace Economics, Peace Science and Public Policy 20 (4). DOI: https://doi.org/10.1515/peps-2014-0038. Easterly, W. and Levine, P. 1997. Africa s Growth Tragedy: Policies and Ethnic Divisions The Quarterly Journal of Economics 112 (4): 1203-1250. Fearon, J. D. 2005. Primary Commodity Exports and Civil War. Journal of Conflict Resolution 49 (4): 483 507. Fearon, J. D., and D. D. Laitin. 2003. Ethnicity, Insurgency, and Civil War. American Political Science Review 97 (1): 75 90. Fearon, J. D., K. Kasara, and D. D. Laitin. 2007. Ethnic Minority Rule and Civil War Onset. American Political Science Review 101 (1): 187 93. Greene, W., 2010. Testing hypotheses about interaction terms in nonlinear models. Economics Letters, 107(2): 291-96. Grossman, H. I. 1991. A General Equilibrium Model of Insurrections. American Economic Review 81 (4): 912 21. Hall, D. B. 2000. Zero-inflated Poisson and Binomial Regression with Random Effects: A Case Study. Biometrics 56 (4): 1030 9. Harris, M. N., and X. Zhao. 2007. A Zero-inflated Ordered Probit Model with an Application to Modelling Tobacco Consumption. Journal of Econometrics 141 (2) :1073 99. Hegre, H., T. Ellingsen, S. Gates, and N. P. Gleditsch. 2001. Toward a Democratic Civil Peace? Democracy, Political Change, and Civil War, 1816 1992. American Political Science Review 95(1): 33 48. Hendrix, C. S., and S. M. Glaser. 2007. Trends and Triggers: Climate, Climate Change and Civil Conflict in Sub-Saharan Africa. Political Geography. 26: 695 715. 22