Horizontal Inequalities and Civil War

Transcription

1 Horizontal Inequalities and Civil War Do Ethnic Group Inequalities Influence the Risk of Domestic Armed Conflict? Gudrun Østby Department of Sociology and Political Science, Norwegian University of Science and Technology (NTNU) & Centre for the Study of Civil War, International Peace Research Institute, Oslo (PRIO) Cand. Polit. Thesis in Political Science August 2003

2 Abstract Recent studies of civil war (notably various studies conducted for the major World Bank study of civil war) conclude that vertical inequality (between individuals) does not increase the risk of internal armed conflict. This thesis argues that such studies neglect a vital aspect of human well-being and conflict the group aspect. Systematic inequalities that coincide with ethnic cleavages in a country are referred to as horizontal inequalities. Furthermore, to grasp the complexity of the inequality conflict nexus we must also look at various dimensions of inequality, not only the economic one. Theoretically, inequality is generally regarded as a typical grievance factor. This thesis argues that horizontal inequality may also be an opportunity factor, and proposes a synthetic theory of horizontal inequality and conflict. Using data from the Demographic and Health Surveys (DHS) for 33 developing countries, this thesis reports aggregated macro-indicators on inequality between ethnic groups, along three dimensions: social, economic and health-related. The thesis analyzes the relationship between such horizontal inequalities among the two largest ethnic groups in a country and their effects on the incidence and onset of domestic armed conflict, respectively. The main finding of this study is that horizontal inequalities between ethnic groups seem to be positively related to domestic armed conflict. Furthermore, I find some evidence that vertical inequality is negatively related to domestic armed conflict. This indicates that horizontal and vertical inequalities have opposite effects on civil war, and that societies with sharp inequalities between ethnic groups, but with low inequality within the groups have the highest risk of domestic armed conflict.

3 Horizontal Inequalities and Civil War Do Ethnic Group Inequalities Influence the Risk of Domestic Armed Conflict? Gudrun Østby Department of Sociology and Political Science, Norwegian University of Science and Technology (NTNU) & Centre for the Study of Civil War, International Peace Research Institute, Oslo (PRIO) Cand. Polit. Thesis in Political Science August 2003

4

5 Acknowledgements Writing this thesis, I have benefited greatly from my affiliation with two excellent research environments: the Department of Sociology and Political Science at the Norwegian University of Science and Technology (NTNU) and the International Peace Research Institute, Oslo (PRIO), where my study has been associated with the Conditions of War and Peace Programme and the Centre for the Study of Civil War. First and foremost, I am grateful to my supervisor, Tanja Ellingsen at NTNU, for her supervision beyond the call of duty in terms of quality, encouragement and time(!). Tanja has been constructive and supportive throughout the whole process, even during her summer vacation. The same goes for my second supervisor, Nils Petter Gleditsch at PRIO/NTNU. I am very thankful to him for providing me with ideas, inspiration, comments and funding, for giving me the opportunity to present parts of my work to a number of scholars at PRIO and elsewhere and, not least, for introducing me to Håvard Strand Others who have provided useful comments, ideas and methodological assistance along the road, include Halvard Buhaug, Tove Grete Lie, Henrik Urdal, Håvard Hegre, Kristian S. Gleditsch and Arild Blekesaune. Earlier drafts of the thesis were presented to the VIP seminars at the Department of Sociology and Political Science at NTNU and to the National Conference on Political Science, Trondheim, 8 10 January The comments made by participants on these occasions have been most fruitful. A student scholarship from PRIO eased my financial situation during this period. It was funded by the Conditions of War and Peace Programme from 1 September 2002 and by the Centre for the Study of Civil War from 1 January I would also like to thank my friends and co- spirrevips, Ragnhild Nordås and Helga M. Binningsbø, for useful suggestions as well as delicious dinners; Robert Wiik for proofreading and especially Birgitte Sivertsen for night-long discussions, and for being such a good friend! My parents and my brother also deserve thanks for taking such great interest in my work and for language editing and moral support. Above all, I thank Håvard for all kinds of advice, for extensive methodological assistance and for his enthusiasm and encouragement! Despite the generous support from all the above-mentioned people, remaining errors are solely my responsibility. Dragvoll, 15 August 2003 Gudrun Østby

6

7 Contents 1 INTRODUCTION THE INEQUALITY CONFLICT NEXUS: THE STATE OF THE LITERATURE Greed or Grievance? Previous Empirical Results Problems with the Inequality Conflict Literature Methodological Objections Conceptual and Theoretical Objections Summary of the Literature. Where Do We Go From Here? THEORY OF HORIZONTAL INEQUALITIES Horizontal Inequalities Defined Why are Horizontal Inequalities Relevant to Violent Conflict? Identity Individual vs. Group Identity Ethnic Identity and Conflict Indicators of Ethnic Diversity Frustration and Relative Deprivation General Theories of Relative Deprivation and Conflict Galtung s Structural Theory of Relative Deprivation Individual vs. Collective Relative Deprivation Indicators of Individual Relative Deprivation: Vertical Inequalities Indicators of Collective Relative Deprivation: Horizontal Inequalities Opportunities for Mobilization Internal vs. External Opportunities An Indicator of Internal Opportunity: Consistency of Horizontal Inequalities Toward a Synthetic Theory of Horizontal Inequalities and Conflict A Complex Relationship: Intra-Group vs. Inter-Group Inequalities Summary of Hypotheses HOW TO MEASURE HORIZONTAL INEQUALITIES Previous Attempts Possible Data Sources and Measures Improving the Minorities at Risk Dataset Combining Data on Vertical Inequality and Ethnic Fragmentation Aggregating Data from the Demographic and Health Surveys RESEARCH DESIGN Qualitative vs. Quantitative Approaches The Sample: Coverage in Time and Space Operationalizations Dependent Variable(s): Domestic Armed Conflict Culture and Identity Vertical Inequalities Horizontal Inequalities Control Variables Controls for Statistical Dependency Reliability and Validity of Data... 74

8 5.5 Methodological Challenges Statistical Method: Logit Regression Data Summary EMPIRICAL ANALYSIS Testing the Inequality Hypotheses on Armed Conflict Incidence Ethnic Identity, Individual Relative Deprivation and Conflict Collective Relative Deprivation and Conflict Inter-Group and Intra-Group Inequalities Testing the Inequality Hypotheses on Armed Conflict Onset Ethnic Identity, Individual Relative Deprivation and Conflict Collective Relative Deprivation and Conflict Inter-Group and Intra-Group Inequalities Summary of the Main Findings CONCLUSION Value Added to the Inequality Conflict Literature Implications for Future Research Policy Recommendations BIBLIOGRAPHY APPENDICES Appendix A. Countries, Periods and Domestic Armed Conflicts Included Appendix B. Armed Conflicts : DHS-Cases vs. Zero Cases Appendix C. Data on Ethnic Diversity and Horizontal Inequalities Appendix D. Various Characteristics of Ethnic Groups, DHS Appendix E. Calculating Horizontal Economic Inequality: An Example Appendix F. Regressions With and Without Lagged Conflict Year Appendix G. Regressions With and Without Zero Cases Appendix H. Alternative Controls For Time Dependency, Conflict Incidence Appendix I. An Alternative Approach to Armed Conflict Incidence Appendix J. Example of SPSS Syntax from DHS Benin, Appendix K. Correlation Matrices

9 Tables Table 5.1 Descriptive Statistics for All Dependent and Independent Variables Table 6.1 Bivariate Analyses of Armed Conflict Incidence , Developing Countries.. 81 Table 6.2 Risk of Armed Conflict Incidence by Ethnic Diversity and Vertical Inequality Variables , Developing Countries Table 6.3 Risk of Armed Conflict Incidence by Horizontal Inequalities , 43 Developing Countries...92 Table 6.4 Risk of Armed Conflict Incidence by Consistency of Horizontal Inequalities , Developing Countries Table 6.5 Risk of Armed Conflict Incidence by Vertical Income Inequality and Horizontal Economic Inequality Combined , Developing Countries Table 6.6 Bivariate Analyses of Armed Conflict Onset , Developing Countries Table 6.7 Risk of Armed Conflict Onset by Ethnic Diversity and Vertical Inequality Variables , Developing Countries Table 6.8 Risk of Armed Conflict Onset by Horizontal Inequalities , Developing Countries Table 6.9 Risk of Armed Conflict Onset by Consistency of Horizontal Inequalities , Developing Countries Table 6.10 Risk of Armed Conflict Onset by Vertical Income Inequality and Horizontal Economic Inequality Combined , Developing Countries Table 6.11 Evaluation of Hypotheses Figures Figure 3.1 Theoretical Expectation Model of the Effects of Ethnic Identity and Vertical Inequalities on Domestic Armed Conflict Figure 3.2 Theoretical Expectation Model of the Effects of Horizontal Inequalities on Domestic Armed Conflict Figure 3.3 Bredel s Theoretical Model of Types of Inequalities and Systematically Increased Potentials for Conflict Figure 4.1 Four Different Combinations of Horizontal and Vertical Inequalities Figure 5.1 Armed Conflict Incidence and Conflict Onsets in 43 Developing Countries, , Relative to the Number of Countries per Year Figure 6.1 Probability of Armed Conflict Incidence in Developing Countries as a Function of Vertical Income Inequality, All Control Variables at Mean Figure 6.2 Probability of Armed Conflict Incidence in Developing Countries as a Function of Regime Type, All Control Variables at Mean Figure 6.3 Comparing the Conflict Potential of Three Dimensions of HIs Figure 6.4 Probability of Armed Conflict Incidencein Developing Countries as a Function of Economic HI, All Control Variables at Mean (and Social HI Removed from the Model) Figure 6.5 Probability of Armed Conflict Incidence in Developing Countries as a Function of Regime Type and Horizontal Social Inequality, All Control Variables at Mean Figure 6.6 Probability of Armed Conflict Incidence in Developing Countries as a Function of Horizontal AND Vertical Inequalities, All Control Variables at Mean Figure 6.7 My Empirical Results of Inequalities and Increased Potentials for Conflict, Compared with Bredel s Theoretical Model

10

11 1 Introduction simple inequality between rich and poor is not enough to cause violent conflict. What is highly explosive is horizontal inequality: when power and resources are unequally distributed between groups that are also differentiated in other ways for instance by race, religion or language. So-called ethnic conflicts occur between groups which are distinct in one or more of these ways, when one of them feels it is being discriminated against, or another enjoys privileges which it fears to lose. (Kofi Annan, 1999) Armed conflicts entail heavy human, economic and social costs and are a major source of underdevelopment and poverty around the world. Civil war is today far more common than interstate conflict. Of the 34 armed conflicts listed in the Armed Conflict Dataset (Gleditsch et al., 2002) for 2001, all but one were internal. Armed conflicts occurred in 28 different countries, and in 11 of these conflicts more than 1,000 people were killed (ibid). An increasing share of intrastate wars have been fought between ethnic groups (Fearon & Laitin, 2002a), for example the civil wars in Uganda and Sri Lanka. Despite ethnicity s alleged effect on internal conflict (e.g. Ellingsen, 2000), civil war cannot be caused by ethnic plurality alone. There are many examples of pluralistic societies where ethnic groups live peacefully side by side. Tanzania, for example, has largely avoided the severe internal conflicts of many of its neighbours. In Switzerland, Costa Rica and Uruguay, too, multiple groups have lived together without serious tension. Why do certain multiethnic countries experience civil war, while others do not? To answer this question one needs to go beyond the sheer cultural differences between ethnic groups. Among the numerous ethnic groups in Tanzania, only a few face systematic disadvantages or discrimination (Minority Rights Group International, 1997: 517). As opposed to this, belligerent countries like Uganda and Sri Lanka are infested with persistent socio-economic inequalities between the different ethnic groups. In Uganda average incomes are about twice as high in the South, which is predominantly Bantu, as in the rest of the country, and social services are substantially better (Stewart, 2000a: 18). In Sri Lanka, university admission policies have been tilted against Tamils (Gurr, 2000: 109). According to Murshed & Gates (2003: 3) well-defined collective grievances are required for ethnically based conflict, and these must be seen and believed to be the case. Related to this, Hauge (2003: 65) argues that inequalities that coincide with ethnic cleavages may enhance both grievances and group cohesion among the relatively deprived, and thus 1

12 facilitate mobilization for conflict. It is therefore my assumption that an important factor that differentiates the violent from the peaceful multiethnic societies is the existence of severe, systematic inequalities between ethnic groups. I refer to such inequalities as horizontal inequalities. Traditional studies of inequality generally focus on the individual. Measures of inequality typically relate to the ranking of individuals (or households) vertically within a country, or sometimes the globe. Such inequality I refer to as vertical inequality. The concept of horizontal inequality (HI) has a strong advocate in Frances Stewart (Chair of the International Development Centre at Oxford University). There have been many efforts to apply this term in practice. During the last couple of years it has constantly been invoked at the United Nations, and in his speeches, UN Secretary-General Kofi Annan has often linked the sources of domestic conflict to ethnic group inequalities, or so-called horizontal inequalities (Annan, 1999, 2001a,b). 1 Despite efforts of putting horizontal inequalities on the UN agenda, little systematic research has been carried out on the subject. Rather than assessing HIs, most studies of the Inequality Conflict nexus have tended to focus on vertical inequality. As for the latter, a body of quantitative aggregate studies executed over the last 40 years has failed to demonstrate any consistent relationship between vertical inequality and conflict, and has even raised doubts as to whether any strong relationship actually exists. Recent studies of civil war (e.g. Collier & Hoeffler, 2002a; Hegre, Gissinger & Gleditsch, 2003) conclude that (vertical) inequality within a country does not increase the risk of internal armed conflicts. Stewart (2002a: 2) argues that such studies of inequality neglect a vital dimension of human well-being and of social stability, namely the group dimension. Amartya Sen (1992: 117) also argues that general analyses of inequality must, in many cases, proceed in terms of groups rather than specific individuals and that one should focus on inter-group variations. Even though an individual may feel frustrated if he is disadvantaged e.g. economically or politically compared to other individuals in society, he will not start a rebellion on his own. Civil wars are organized group conflicts: that is to say they are not a matter of individuals randomly committing violence against each other. What is involved is group mobilization, and we need to explore the underlying motivation for such conflicts (Stewart, 2000a). Hence, an individualistic approach to measuring inequality does not seem the best starting point for understanding the relationship between inequality and conflict. 1 The link between the academic and practical approaches to horizontal inequalities and civil war has been the World Institute for Development Research in Helsinki, which is a part of the United Nations University (UNU), and it has also been involved in common research activities with Stewart s Center. 2

13 Moreover, the vast majority of inequality conflict studies concentrate exclusively on economic inequality (usually operationalized as income inequality or inequality in land distribution). Stewart (2002a,b) argues that not only is it important to be aware of the group boundaries of horizontal inequalities, but that inequality is also multidimensional. That is, in order to embrace the whole picture of the inequality conflict relationship, inequalities should be measured along various dimensions, not only the economic one. In line with this, Sen (1992: 28) argues that an important and frequently encountered problem arises from concentrating on economic inequality as the primary focus of attention in the analysis of inequality. According to Stewart (2002a) horizontal inequalities can be divided into three broad dimensions: political, economic and social, with each category containing a number of variables. For example, social inequality may occur at the level of access to education and employment. Based on nine case studies, Stewart found that various dimensions of horizontal inequalities provoked some kind of conflict, ranging from a high level of criminality in Brazil to civil war in Uganda and Sri Lanka. The lessons derived from these case studies are of course valuable, providing deep insight into the specific cases. However, making generalizations as to the universal relationship between horizontal inequalities and violent conflict based on few cases is problematic. For this purpose, a systematic, quantitative research design is needed. The aim of this thesis is to take into consideration the critics of Stewart (2000a, 2002a,b) and Sen (1992) in order to clarify whether various dimensions of horizontal inequalities affect the probability of domestic armed conflict, when tested quantitatively across many cases. My main research hypothesis is the following: Countries with severe horizontal inequalities are more likely to experience domestic armed conflict than countries with weak or no horizontal inequalities. Theoretically, previous research on civil war has identified three main preconditions of group mobilization: identity, frustration and opportunity (e.g. Ellingsen, 2000). Despite the fact that there seems to be considerable overlap between social identity theory, relative deprivation theory and theories of mobilization opportunities, theoretical specifications and empirical studies mainly point to the distinctive contribution of each perspective. This thesis tries to achieve a more integrative theoretical analysis in which these perspectives are explicitly related to each other. A common identity is innate in the concept of horizontal inequalities, as ethnic identity groups are the main focus. Furthermore, inequality is generally 3

14 considered as a typical frustration factor. This also applies to horizontal inequalities because membership in relatively deprived groups can cause resentment among individuals on behalf of the group. This thesis argues, however, that the sum of horizontal inequalities (i.e. their consistency across various dimensions) may as well affect the opportunities that a group has to mobilize. Thus, I propose a theory of horizontal inequalities and civil war that synthesizes all the three factors above identity, frustration and opportunity. On the empirical side, this thesis provides new data and a measure of horizontal inequalities quite different from other measures in the literature. I construct aggregated macro-indicators for group differences based on survey data that include information about economic and social indicators, as well as ethnic origin. Generating data from the Demographic and Health Surveys, I aggregate measures of horizontal inequalities along three dimensions: economic, social and a third dimension that I have labelled the health-related dimension. By doing this, I am able to measure various types of horizontal inequalities and assess what impact they have on domestic armed conflict, both separately and jointly. The study is restricted to domestic armed conflict because the theories I aim to test relates empirically to civil war. An additional argument for focusing on internal conflict is that this is by far the most frequent type of conflict today, and thus the most serious threat to human security and development. Using the Armed Conflict Dataset, which includes all conflicts with a minimum of 25 battle-related deaths (Gleditsch et al., 2002) I test all my hypotheses with regard to both incidence and onset of internal conflict. Throughout the thesis, the term conflict or armed conflict refers to domestic armed conflict unless otherwise stated. The thesis is structured as follows: In Chapter 2 I review some of the more acknowledged works concerning the relationship between (vertical) inequality and civil war, and identify some methodological and conceptual flaws that hamper most of these studies. In Chapter 3 I present theories on ethnic identity, frustration and opportunities. Based on these, I derive three sets of hypotheses relating to the risk of domestic armed conflict: one concerning the effects of ethnic diversity and vertical inequality; another concerning the impact of horizontal inequalities and a third combining the effects of vertical and horizontal inequalities. The whole of Chapter 4 is devoted to the measurement of horizontal inequalities and where to find data. I argue that the best option is to construct horizontal inequality ratios based on national survey data. In Chapter 5 I present the research design, which includes a methodological presentation and discussion and the operationalizations of my variables. In Chapter 6 I present the empirical analysis, testing my hypotheses with regard to both 4

15 incidence and outbreak of domestic armed conflict. The conclusion in Chapter 7 summarizes the thesis. I describe my contributions to the inequality conflict research, as well as the problematic sides to my thesis, presenting some challenges to future research in the field. Finally, I provide some policy recommendations. The main finding of this thesis is that horizontal inequalities between ethnic groups seem to be positively related to domestic armed conflict. Furthermore, I find some evidence that vertical inequality is negatively related to domestic armed conflict. This indicates that horizontal and vertical inequalities have opposite effects on civil war, and that societies with sharp inequalities between ethnic groups, but with low inequality within the groups have the highest risk of domestic armed conflict. 5

16 2 The Inequality Conflict Nexus: The State of the Literature I find myself perplexed: why do tests of the same hypothesis using different data, research designs, and methodologies appear to produce such dramatically different conclusions? (Zinnes, 1976: 212) At least since Aristotle theorists have believed that political discontent and its consequences protest, instability, violence, revolution depend not only on the absolute level of economic wealth, but also its distribution (Nagel, 1974: 453). A remarkably diverse literature, both ancient and modern, theoretical as well as empirical, has coalesced on the proposition that political violence is a function of economic inequality. So much systematic quantitative evidence has been provided in order to discover the true relationship between (vertical) economic inequality and political conflict that Lichbach (1989: 435) was able to locate 43 aggregate quantitative studies, within nations and cross-national, of the economic inequalitypolitical conflict nexus, and many more have been conducted in the years following Lichbach s review. There are, in theory, five possible relationships between economic inequality and political conflict: positive, negative, convex (inverted U-shaped), concave (Ushaped), or null. Reviewing the empirical literature, we can find examples of all. 2 It has been persuasively argued (e.g. Davies, 1962; Feierabend & Feierabend, 1966; Gurr, 1970) that frustration, or relative deprivation, is a major determinant of political protest and violence. Most studies of inequality and conflict relate somehow to the theory of relative deprivation. This theory argues that while absolute poverty may lead to apathy and inactivity, comparisons with those in the same society who do better may inspire radical action and even violence. As opposed to the theory of relative deprivation, advocates of what has come to be called the resource mobilization - or mobilization opportunity approach to the explanation of collective violence and protest (Tilly, 1978; Collier & Hoeffler, 2002a) reject such frustration hypotheses for the reason that inequality and discontent are more or less always present in practically all societies. Consequently, they believe that the most direct and influential explanatory factors are not perceived grievances, but rather financial and political opportunities for mobilizing a rebel organization. 2 Lichbach (1989: ) identifies seventeen studies which posit that the inequality conflict nexus is positive. He finds only a handful of scholars who anticipate that the relationship will be negative. As regards curvilinear relationships, Lichbach identifies six studies, four of which suggest that it is convex (U-shaped) and two which suggest that it is concave (an inverted U-shape). Finally, several of the studies fail to find a significant inequality conflict nexus at all. 6

17 For almost four decades, political analysts from both traditions have tried to test the assumption that inequality breeds political conflict, using both qualitative case studies as well as modern statistical methods. But despite the persistence of this theme, the literature on inequality and political violence has not produced a conclusive answer to the question What is the relationship between inequality and civil war? This literature review proceeds as follows: I first review some of the most important empirical studies of vertical inequality and conflict, starting with Collier & Hoeffler s recent, but well-known dichotomy of greed and grievance in civil war. Secondly, I discuss some of the problems associated with previous studies on vertical inequality and conflict. Finally, I summarize the literature, and suggest how a reconceptualization of inequality might be a solution to the empirical confusion in the field. 2.1 Greed or Grievance? A recent trend in theories of civil war is to emphasize the economic or other material benefits that elites stand to gain from civil war. Both economists and political scientists have argued that financial motives and constraints may be the determinants of civil war (Sambanis, 2002). A theoretical distinction has been drawn between so-called greed-driven and grievancedriven rebellion. The main advocates of this dichotomy are Paul Collier & Anke Hoeffler from the World Bank. In several papers (Collier & Hoeffler, 2002a,b; Collier, 2000) they have discussed whether civil conflicts are caused by greed or grievance. They present two alternative explanations for civil war: atypical grievances or atypical opportunities for forming a rebel organization. Collier & Hoeffler s grievance model consists of factors such as high income inequality, a lack of political rights and ethnic and religious divisions in society. Among the greed factors in their opportunity model, are access to finance, such as the scope for extortion of natural resources, and geographical factors such as the extent of mountains and forests. Using a global data set of civil wars for the period they test the two explanations, and find that opportunity provides considerably more explanatory power than grievance. Inequality is among the grievance factors largely dismissed by Collier & Hoeffler. Based on data on income inequality generated by Deininger & Squire (1996), inequality has no significant effect on civil war in their grievance model. Even though Collier & Hoeffler (2002a: 1) claim that they do not actually reject one model in favour of the other, they conclude by saying that greed considerably outperforms grievance. Critics (e.g. Herbst, 2000; Kalyvas, 2001) argue that the greed-grievance distinction may not be as clear as the proponents of economic theories of civil war might suggest. 7

18 According to Kalyvas (2001: 104), the main problem is the direction of causality: do people wage war in order to loot or do they loot to be able to wage war? What he is arguing, then, is that one cannot normally distinguish between war loot that simply serves as a means to sustain the war effort and loot that is the ultimate aim of war (greed). Moreover, it is not always clear who is doing the looting (e.g. elites, autonomous militias or armed peasants). Finally, according to Kalyvas, the linkages between opportunities and grievances are complex and fluid. There are a number of technical reasons why a relationship between inequality and conflict might not be supported by Collier & Hoeffler s model, such as lacking control variables and missing data. One of the most important reasons, however, may be conceptual. Whereas Collier & Hoeffler test vertical inequality (inequality between individuals), case study work suggests that it is not necessarily inequality between individuals that matters in causing conflict, but rather inequality between ethnic groups (horizontal inequality). Econometric work that uses measures of overall inequality fails to capture these effects, and this might be a reason why Collier & Hoeffler s grievance model does not obtain as much explanatory power as it might with an alternative measure of horizontal inequality. The greed-grievance (or opportunity-frustration) debate merits further examination. It is not my intention, however, to frame the debate in either-or terms, as the key seems to be in understanding the interaction between the two. 3 I only argue that the traditional grievance explanations might be underestimated in Collier & Hoeffler s model due to dissatisfactory measurements of and a too individualistic approach to inequality, as is discussed in more detail later in this chapter. Before that, however, I give a brief historical overview of other important studies of the relationship between vertical inequality and political conflict. Most of these studies use relative deprivation as their theoretical framework, considering inequality as a grievance factor, just like Collier & Hoeffler. 2.2 Previous Empirical Results The pioneering cross-national research on the inequality-violence relationship was Russett s (1964) study of the political impact of inequality in land tenure systems. Although he 3 Collier & Hoeffler (2002a: 6) actually admit that rebellions motivated by greed is merely a special case of general opportunity-theories focusing on the various constraints upon rebellion. An alternative constraint-based theory is that of universal grievance: All countries might have groups with a sufficiently strong sense of grievance to wish to launch a rebellion, so that rebellions will occur where they are viable. However, Collier & Hoeffler, a bit confusingly, refer to different theories of rebellion by the shorthand of greed (in order to contrast them with preference-based theories, like that of relative deprivation). Hence, when Collier & Hoeffler are talking about greed, one cannot be sure whether they are actually referring to greed as such. 8

19 expected to find strong relationships between national levels of land inequality and political instability, Russett could document only low to moderate correlations between several indicators of each in 47 countries. Thus, Russett was forced to conclude that the positive relationship between inequality and conflict was only moderate. A subsequent study by Parvin (1973) provided even less support for the idea that political violence is a function of economic inequality. Working with a sample of 26 predominantly Western nations, he regressed a measure of political unrest on a set of economic determinants, like intersectoral income inequality. The contribution of inequality proved to be only marginally significant and even more surprising to Parvin inverse. That is, political unrest was found to occur with greater frequency in more egalitarian societies. Parvin could not adequately account for this unanticipated finding. 4 Since arguments and evidence had been offered for the existence of both a positive (Russett, 1964) and a negative (Parvin, 1973) relationship between inequality and conflict, it is not surprising that researchers have tried to combine the two assumptions and resolve the contradiction. One of them is Nagel (1974). His research results did not exactly clarify matters. For Nagel, the key to political violence was to be found in relative deprivation. The discontent triggered by inequality, Nagel believed consisted of the tendency of individuals to compare wealth (a tendency Nagel believed was inversely related to the amount of objective inequality), and the extent of the grievance resulting from such comparisons (a direct function of inequality). Combining these two factors multiplicatively, Nagel suggested that the inequality-violence relationship resembled an inverse U-curve, with political violence most likely at intermediate levels of inequality. Although his analysis of Vietnamese provinces produced results that were in accordance with his curvilinear assumption, Nagel s attempt to replicate these findings over a cross-national sample was unsuccessful. 5 Sigelman & Simpson (1977) were the first to have access to personal income data. Their sample consisted of 49 nations, comprising almost all of those for which data on personal income inequality were available at the time. They assumed that data on personal income inequality would have greater violence potential than land inequality because in many societies particularly those at higher development levels life chances are not so closely 4 According to Weede (1981) Parvin s results permit two interpretations. First, with ordinary, unweighted regressions, inequality seems to contribute to less instead of more political unrest. Secondly, if one weighs nations according to population size, one gets the positive sign, but the effect becomes insignificant. Moreover, Weede argues that Parvin s data must be treated with caution because he used sectoral instead of household or personal income distribution data, and that such data necessarily underestimate inequality. 5 Davis (1948) offered another curvilinear interpretation of the inequality-violence relationship, diametrically opposed to that of Nagel. A narrow concentration of incomes, Davis believed, would spur mass resentment and lead to revolution, while a wide dispersion would endanger elite dissatisfaction and ultimately cause civil war. 9

20 connected to land ownership. Relying on the theorizing of Nagel (1974), Sigelman & Simpson investigated the possibility of a curvilinear as well as a linear relationship between the Gini index 6 for personal income inequality and internal war. They provided no support for a curvilinear relationship, but some support for a linear one. When controlling for e.g. GNP per capita, inequality lost some of its predictive power. Therefore, Sigelman & Simpson (1977: 124) concluded that the overall level of societal well-being is a more critical determinant of political violence than is income inequality. These results were supported in a study by Weede (1981), who found that there is a stronger relationship between average income and collective violence than between income inequality and collective violence. In fact, he found no significant relationship between income inequality and violence at all, and thus could not add support to the relative deprivation explanation of political violence. Muller & Seligson s (1987) compared the causal importance of land inequality versus income inequality as determinants of mass political violence cross-nationally, assuming that the latter would be more important. Their hypothesis was based on an integration of relative deprivation theory with the resource mobilization approach: They assumed that whereas a high level of income inequality nationwide would significantly raise the probability that at least some dissident groups would be able to organize for violent collective action, a high level of agrarian inequality would not have the same effect on collective violence because it would be more difficult to mobilize people in the countryside. They found support for a positive relationship between income inequality and domestic conflict. Land inequality, they concluded, was relevant only to the extent that it was associated with the nationwide distribution of income inequality. In 1996 Alesina & Perotti s added support to the theory of relative deprivation. In a cross-sectional study of 71 developing countries they found that income inequality, by fuelling social discontent, increased political instability. Despite the sample period of 25 years ( ), Alesina & Perotti had only one observation of income inequality for each country. The problem of limited temporal domain restricting the samples of most inequality conflict studies was partly overcome later the same year, when Deininger & Squire (1996) presented a 6 The most common measure of income inequality is the Gini coefficient an index between 0 and 1 (or 0 and 100) where 0 implies an egalitarian distribution (perfect equality) and 1 (or 100) indicates total concentration (perfect inequality). Egalitarian countries like Sweden and Japan have Gini scores around Highly unequal countries, like South Africa, have Gini scores around 0.60 and more. The Gini coefficient is defined graphically as the area of concentration between the Lorenz curve and the line of perfect equality. The Lorenz curve is a graphical representation of the proportionality of a distribution (the cumulative percentage of the values). For an introduction, see Lorenz (1905). See also Sen (1997) for a comprehensive overview of inequality measures. 10

21 new data set on income inequality. The new data represented an improvement in quality as well as an expansion in coverage not only temporal, but also geographic. Since the Deininger & Squire data set was introduced, all large N-studies of vertical inequality and conflict that I am aware of have applied these data. Amongst these are Auvinen & Nafziger s (1999) econometric analysis of the relationship between complex humanitarian emergencies 7 and factors such as income inequality. With data from 124 developing countries for the period , their regression analyses indicated that severe income inequality is associated with political conflict and complex humanitarian emergencies. Auvinen & Nafziger (ibid: 278, fn. 5) actually attributed their finding that there is a positive link between inequality and domestic conflict to the Deininger & Squire data set: We were able to find relationships between Gini and war, which other researchers (such as Collier and Hoeffler 1996) could not find without this data set. 8 Interestingly, shortly after Auvinen & Nafziger s article, several researchers of civil war, including Collier & Hoeffler (2002a) using the same inequality data (Deininger & Squire, 1996) concluded that there is no significant relationship between inequality and domestic conflict, hence, that inequality is irrelevant to dissidence. In a structuralist model of armed conflict, Hegre, Gissinger & Gleditsch (2003) tested a two-stage-hypothesis of the relationship between globalization, inequality and armed conflict. They found some evidence that trade increases income inequality. However, like Collier & Hoeffler (2002a), they did not find any relationship between vertical inequality and civil war. 2.3 Problems with the Inequality Conflict Literature There are a number of potential reasons why the studies reviewed here come to so different conclusions with regard to the relationship between inequality and conflict. I have divided the 7 According to Auvinen & Nafziger a complex humanitarian emergency (CHE) is a human-made crisis in which large numbers of people die and suffer from war, physical violence (often by the state), or displacement, and is usually accompanied by widespread disease and hunger (1999: 267). 8 The Collier & Hoeffler paper that Auvinen & Nafziger refer to, On Economic Causes of Civil War was published two years later (1998) in Oxford Economic Papers 50: Quite correctly, Collier & Hoeffler at that point argued that there is insufficient data to introduce distributional considerations into the empirical analysis (1998: 563). Elsewhere, Nafziger & Auvinen (2001: 9) voiced their criticism against another Collier & Hoeffler paper, stating that they use inadequate income inequality data (However, Nafziger & Auvinen obviously did not know what income inequality data Collier & Hoeffler had used). Anyway, in the well-known paper Greed versus Grievance in Civil War, Collier & Hoeffler explicitly did use the Deininger & Squire (1996) income inequality data, just like Auvinen & Nafziger (1999), and they actually found that inequality has no significant effect on civil war. Hence, I assume that the reason for the opposite findings may rather be attributed to the different research designs of the two studies, e.g. in terms of operationalizations of the dependent variable (CHEs vs. civil war), samples (developing countries vs. whole world), and different control variables and methodologies. 11

22 critique into two main parts. The first considers various methodological problems, and the second provides a more fundamental critique, relating to the conceptualization of inequality Methodological Objections It has been argued that the contradictory inequality conflict results are due to variations among the studies in all aspects of research design (Lichbach, 1989). Various critics have suggested that the inconsistent conclusions arise from a lack of essential control variables, from the different cases explored, from the various time frames in which the effects of conflict are examined and not least from poor data and inadequate measurement of inequality. Lichbach (1989) suggests that those studies that find a positive relationship between inequality and conflict may be spurious because they failed to include control variables like the level of economic development and regime type. In line with this, Zimmermann (1983) argues that inequality and level of economic development must be considered simultaneously for adequate analysis of conflict, because economic development may confuse the inequality conflict nexus. Inequality should be curvilinearly related to economic development, reaching its peak at a mid-level of development The interrelationship between these two independent variables used in explaining political violence may account for inequality not showing the predicted relationship. (Zimmermann, 1983: 137) 9 A related critique comes from Hegre, Gissinger & Gleditsch (2003: 257), who claim that scholars have focused on relative deprivation at the cost of ignoring more important explanatory factors. The cases sampled by statistical modelers in inequality conflict studies have varied greatly. Some of the recent cross-national studies have employed a global sample of states (Collier & Hoeffler, 2002a,b; Hegre, Gissinger & Gleditsch, 2003). Others have focussed on a restricted spatial domain. Nagel (1974) studied the relationship between inequality and conflict in 26 Vietnamese provinces, whereas Parvin s (1973) sample consists of 26 predominantly Western nations. While there may be a number of good reasons for doing so, focusing on a limited number of states makes generalisations inappropriate due to a potential lack of representativeness. Another potential flaw affecting the majority of the studies reviewed is their restricted temporal domain. Most of the studies reviewed (e.g. Russett, 1964; Nagel, 1974; Muller & 9 A curvilinear relationship between inequality and economic development was first proposed by Kuznets (1955), and is often referred to as the Kuznets curve. 12

23 Seligson, 1987) are cross-sectional studies, with only one observation for each variable. A cross-sectional study is not the best approach to analysing domestic conflict, which may erupt at any given time during the observation period. Consequently, it makes it problematic to study the relationship between inequality and conflict over time. Finally, one of the most serious objections concerns the poor data on income inequality and the high level of missing observations. 10 Indices of income inequality data are measured with particularly high levels of error (Cramer, 2001). Add to this that for many countries we hardly have any income inequality data at all. Before the Deiniger & Squire (1996) dataset, and the recent appearance of the World Income Inequality Database (WIID) (UNU/WIDER & UNDP, 2000) 11, cross-national data on inequality were distressingly scarce and imprecise. Yet, with this progress, the problem of a very large amount of missing data is still present. Deiniger & Squire include inequality data from a very limited number of countries and years. They use an objective and valid operationalization of income inequality, but, as noted by Székely & Hilgert (1999), the observations are not always comparable because they do not necessarily refer to the same notion of income. 12 Cramer (2001) provides a critical analysis of the general economic literature on inequality and conflict. He is sceptical to any multinational study of the inequality conflict nexus: Whatever the apparent consensus in the field at any particular moment it can never be stressed enough how weak the empirical foundations remain for any construction based, especially, on inter-country comparisons. (Cramer, 2001: 2) Strand & Gates (2002) argue that worse than the problem of missing data itself, is the problem that arises when the pattern of missing data is non-random, or biased. Many countries do not have any inequality data at all. Trying to locate income inequality for these countries, Strand & Gates put a request to the Scientific Study of International Processes (SSIP) listserv 13 for information, and got the following answer from Phil Schrodt (quoted in Strand & Gates, 2002: 5 6): 10 In Chapter 5 I show that some of the problems mentioned above also affect the study presented in this thesis. 11 The latter builds on Deininger & Squire s dataset, which forms half of the current WIID. 12 Also, the observations vary with regard to the unit of observation (household vs. individual) and population coverage (e.g. all country vs. capital) (Deininger & Squire, 1996). 13 SSIP is dedicated to bringing together researchers who, at all levels of analysis and with respect to the entire range of international political questions, pursue issues using (1) formally stated arguments and/or (2) systematically collected and analyzed empirical data. Eligibility for participation in SSIP requires membership in ISA and formal affiliation with the ISA-SSIP section (SSIP homepage: 13

24 Missing data is usually missing for a reason and this is a splendid example. Seems to me almost all of these cases fall into one of three cases: 1. None of your business, infidel; 2. None of your business, capitalist running dog CIA lackey; 3. We d be delighted to give you the information, but we haven t had a decent meal in thirty years; 4. All of the above (Somalia). When studying conflict, a biased sample of cases may disturb the effect of the independent variable. Humphreys (2002: 3) notes that if we are less likely to have income inequality data for countries where there are civil wars, this will bias the estimated effect of inequality downwards, which will make us believe that the relationship is weaker than it is. Cramer (2001) provides a further criticism of the traditional vertical inequality data. He is particularly sceptical to the Gini coefficient, which he describes as a superficial outward sign of inequality. This point is illustrated by the examples of Indonesia and Rwanda, both of which are commonly regarded as two countries with low Gini coefficients. Cramer claims that to draw from the published data on inequality that either of these two countries is a low-inequality country would be misleading, or even absurd: Indonesia has probably experienced rapid increase in income and wealth inequality in recent years, a fact that is directly observable to the eye in and around Jakarta, for example, with its extravagant shopping emporia coexisting with extreme poverty and, further afield, dire indigence in rural areas. Rwanda also is not quite the Cuba or Kerala of its Gini image (Cramer, 2001: 5 6). Cramer is making the point that there may be severe inequalities locally (micro-level) even though a country, on the whole, scores relatively low on the Gini index. He claims that in the majority of civil conflicts the intensity of violence is conflict at close quarters, i.e. about visible and felt inequalities at the local level rather than the extremes of the Gini coefficient and the ratio between earnings of the richest and the poorest quintile of the population. In Rwanda, there is also sharp categorical inequality, 14 between ethnic groups (Cramer, 2001: 12), a fact that is not captured by the traditional vertical inequality data due to the country s low Gini score. 15 What matters, according to Cramer, is how inequality is institutionalised and shaped by history and various social and cultural cleavages. This last point touches upon a more fundamental critique of the previous inequality conflict studies which is elaborated more in detail below. 14 Whereas Cramer discusses categorical inequality, I prefer the term horizontal inequality. The two concepts, however, mean the same thing, and might be used interchangeably. 15 Rwanda s Gini coefficient in was (UNI/WIDER & UNDP, 2000), which is a rather low value. 14