Chapter

4 The Economics of Civil War in Sub-Saharan Africa

Author(s):
Jean Clément
Published Date:
February 2005
Share
  • ShareShare
Show Summary Details
Author(s)

Section I. Introduction

We have always maintained that the conflict in Sierra Leone is not about ideological, tribal or regional differences. It has nothing to do with the so-called problem of marginalized youths or, as some political commentators have characterized it, an uprising by rural poor against the urban elite. The root of the conflict is diamonds, diamonds, and diamonds.

Ibrahim Kamara, Sierra Leone Ambassador to the United Nations1

Low-intensity conflicts, civil strife, and war are major threats to economic growth and sustainable development. Globally, there have been about 140 civil wars since the end of World War II. These wars have killed approximately 20 million people and displaced about 67 million (Doyle and Sambanis, 2003). In sub-Saharan Africa, about 20 countries have experienced at least one period of civil war since independence. The situation is particularly dire for countries such as Angola and Sudan, which have hardly experienced any significant period of peace since independence. While the incidence of civil war has decreased globally, the incidence and intensity of civil war in Africa have been on the rise in the past few decades. According to the Stockholm International Peace Research Institute (SIPRI) Yearbook 2000, of the 27 active armed conflicts going on around the world in 1999, about 41 percent were civil wars taking place in Africa.

The adverse effects of civil wars on economic growth and development can hardly be overemphasized.2 The World Bank has labeled civil wars as “development in reverse”3 because of the reversal effects of civil wars on economic development.4 The risk of civil war is much higher in low-income countries than in middle-income or high-income countries. Civil wars, therefore, reflect not only a problem for development but also a failure of development (Collier and others, 2003). Between 1960 and 1999, the per capita income of countries affected by civil wars, on average, was less than that of countries that sustained peace. In addition, countries that sustained peace, on average, grew faster than countries affected by civil wars.5

The forces involved in these civil wars are extremely complex, each situation having its unique patterns of interests, power and powerlessness, devastation, and prosperity. Conflict prevention and peace building cannot be based on unstable solutions. Identifying the root causes of these civil wars is, therefore, crucial in any peace-building efforts. At the very basic level, internal conflicts may stem from oppression, inequality, and ethnic hatreds that manifest themselves in terms of nationalism, separatism, or fight for an ethnic identity (James and Goetze, 2001). In more complex situations, it may be rooted in greed (the desire for economic and financial gains) as opposing interests compete for resource wealth (Collier, 2000a; Grossman, 1999; Tily, 1978). The latter is referred to as the “economic theory of conflict.”

Even though literature on the economics of civil war has increased understanding about economic factors that make a country prone to civil war, little attention has been paid to the specific case of sub-Saharan Africa.6 Most of the studies on Africa are based on inferences from global samples. As Ali (2000) points out, drawing conclusions on the basis of global samples, though useful, is likely to ignore vital regional and country-specific issues that have to do with a complex interaction of socioeconomic, political, and historical legacies. This chapter takes up this challenge by looking specifically at the case of sub-Saharan Africa. The analysis applies the Collier and Hoeffler (C-H) model to sub-Saharan Africa using a panel data set of 44 countries in this region. The objective is to determine which factors are more important in explaining the risk of civil war in the region. More specifically, the chapter examines the extent to which economic growth and the development of good-quality institutions can help prevent civil wars in sub-Saharan Africa.

The C-H analysis is extended by introducing measures of institutional quality7 and corruption. The institutional quality is introduced because many economists believe that high-quality institutions, such as rule of law, democratic accountability, and bureaucratic quality can mitigate any adverse effects of ethnic dominance or ethnic fractionalization identified by Easterly and Levine (1997). In particular, high-quality institutions can help minimize war casualties on national frontiers and even reduce the probability of genocide for any given amount of ethnic fractionalization (Easterly, 2000). The chapter investigates Rodrik’s (1999) assertion that ethnic conflict is not a product of social or religious fractionalization but rather the failure of most developing countries to build institutions to mediate conflicts that arise in any society.

The remainder of the chapter is organized as follows. Section II is a theoretical and empirical discussion of the economic theory of conflict. Section III documents the trends and characteristics of civil wars in sub-Saharan Africa. Section IV presents the empirical model showing the relationship between the probability of conflict and economic, institutional, social, and demographic factors. Section V follows with the empirical results, and Section VI concludes the chapter by prescribing policy guidelines for conflict prevention and peace building in sub-Saharan Africa.

Section II. Economic Theory of Conflict

Theoretical Background

The literature on the causes of internal conflicts can be traced as far back as the nineteenth century.8 As early as 1835, de Tocqueville considered inequality as the fundamental cause of internal conflict. Lichbach (1989) also points out that many revolutions have been based on moving away from repressive regimes and achieving egalitarian ideas.9 Two competing models usually link inequality with the risk of violent conflict: economic discontent theory (Gurr, 1970) and political opportunity theory (Tily, 1978). According to the economic discontent theory, income inequality is the basis of all rebellions, and if economic inequality is high, violent political conflict will certainly occur. Political opportunity theory maintains that economic discontent is not as important and that political resources and opportunities determine the extent of violent political conflict within countries.

The economic theory of civil war (Collier, 2000a; Fearon and Laitin, 2000; Grossman, 1999; Tily, 1978) views civil war as the outcome of an expected utility maximization decision. Rebels are rational individuals who will evaluate the expected benefits from war against the expected costs. Rebellion is, therefore, a rational decision, and the financial viability of the rebellion is what determines whether a rebellion will be carried out. Utility is maximized by starting a rebellion if the gains from winning the war outweigh the costs of coordinating a rebellion and the likelihood that the government will be able to sustain a massive military effort to contain or put down the rebellion. Collier and Hoeffler model the demand for rebel labor as the outcome of underlying grievance and the supply of labor as the result of expected utility maximization. As per capita income rises, the government’s ability to defend itself also increases, and so does the opportunity cost of the rebellion.

The upshot of Collier and Hoeffler’s analysis is that wars in developing countries have become less ideological and are principally fueled not by grievance but by greed, the basic assumption of neoclassical microeconomics. Collier (1999, p. 1) argued that “group grievances beneath which inter-group hatreds lurk, often traced back through history” are not significant factors that make a country prone to war. Instead, economic agendas and economic opportunities are far more likely than social or group grievances to cause civil wars. Collier, defending his analysis, argued that justice, revenge, and relief from grievance are public goods and are therefore subject to free-rider problems that are a disincentive for a rebellion. In addition, people are unwilling to fight for a cause unless they are convinced that the rebellion will succeed; hence, initially, rebellions face a coordination problem. Furthermore, there is a time-consistency problem in that potential recruits can recognize that a leader promising to alleviate grievances may, once in power, turn out not to deliver. However, if a rebellion is motivated by greed, it allows the participants to restrict the benefits to themselves and thereby avoid any free-rider problems.

Critics argue that there is little sense in conflicts feeding on both self-interest and concern for the public good at the same time. Furthermore, the complexity of causes and motivation for a rebellion cannot be captured adequately by the grievance variables, which, as empirically questionable measures of attributes of difference or stratification, are crude tools for capturing social relations in their diversity (Cramer, 1999). In addition, Keen (1998) suggests that some economic motivations for participating in conflict, and indeed for perpetuating and sustaining conflict, may only become paramount once a rebellion has already begun rather than being the main deciding factor in starting a rebellion.

Empirical Evidence

In the economic theory of conflict, economic motivation is paramount. Economic motivation is proxied through the measures of the primary commodities export–GDP ratio, male enrollment in secondary schools, and economic growth. Grievance is proxied by indices of social fractionalization, inequality in land ownership, and an index of political right. Using data for the period 1960–99 and probit and logit techniques, Collier and Hoeffler (2002a) find that social fractionalization, initial income, dependence on primary commodity exports, and population size are strong determinants of the probability of civil war. Collier’s statistical results suggest that some countries are prone to civil war “simply because they offer more inviting economic prospects for rebellion” (Collier, 1999). Collier and Hoeffler conclude that greed is more important than grievances in explaining civil wars.

Fearon and Laitin (2000) also argue that the determinants of conflicts are mainly economic and not political. They test their theoretical model and find that nationalism and cultural cleavages are not important in explaining the prevalence and magnitude of civil wars. Their findings reveal that higher levels of economic development tend to reduce the risk of civil war. Thus, they argue further that civil war will occur when the economic opportunity costs are low and that lack of democracy and ethnic fractionalization are not significant factors in explaining the risk of civil war. Elbadawi and Sambanis (2000b) apply a variant of the Collier-Hoeffler model to analyze the case of Africa.10 They define incidence of civil war as the sum of the probability of war initiation in a period given the presence of peace in the previous period and the probability of war in a period given the presence of war in the previous period. They confirm the Collier-Hoeffler results of the incidence of civil war.

The economic theory of civil war is supported by studies using historical data. Flanagan and Fogelman (1971) studied 65 nations from 1800 to 1960, and conclude that there is less likelihood of civil war breaking out in countries where the levels of economic development are high. Jacobsen (1996) finds no civil wars at all in the period 1945–85 in countries with high levels of economic development. A good explanation for this relationship is that more advanced countries have a higher standard of living and a more highly educated population, along with lower unemployment rates. They are less prone to civil wars because the opportunity cost of a rebellion is very high. The conclusion we infer from the studies reviewed is that a high level of economic development increases the likelihood of domestic peace.

Section III. Civil Wars in Sub-Saharan Africa Defining Civil War

Defining what exactly constitutes civil war is difficult, especially at the level of cross-country analysis, because it is often difficult to distinguish between the beginning and the end of a period to be classified as a war period. Furthermore, all wars are not the same because their causes, intensity, geographical spread, duration, and military characteristics are different (Cramer, 1999). Consequently, there are different views about what should constitute a civil war for the purpose of empirical analysis. The definition of civil war that we adopt in this chapter is derived from Singer and Small’s Correlates of War Project. Singer and Small (1993) define an armed conflict as a civil war if four conditions are met:

  • a major battle took place entirely within the borders of a country,

  • the government is a major combatant,

  • effective resistance occurred on both sides, and

  • at least 1,000 deaths occurred during the course of the war.

Trends and Characteristics of Civil Wars in Africa

According to SIPRI, Africa is the most conflict-ridden region of the world and the only region in which the number of armed conflicts is on the increase. The balkanization of Somalia, the implosion of Sierra Leone, ethnic genocides in Burundi and Rwanda, and the regionalization of rebellion in the Democratic Republic of the Congo (DRC) are some of the violent conflicts that have afflicted Africa since the beginning of the 1990s. Even more striking is that the incidence of war has increased in the past two decades in Africa, while it has fallen in other regions. In 1996, 14 of the 53 countries of Africa were affected by armed conflicts, accounting for more than half of all war-related deaths globally and resulting in more than 8 million refugees, returnees, and displaced people (Annan, 1998).

Table 4.1 lists Africa’s major wars since 1960. Most of the wars in Africa are in the form of civil wars. Wars in Africa are relatively very short on the average and they tend to be among the bloodiest (Elbadawi and Sambanis, 2000b). For the most part, these conflicts are virtually internal conflicts, with some exceptions, including the conflict between Ethiopia and Eritrea. However, in most of these conflicts, there has been massive direct involvement of external actors (Cramer, 1999). In other situations, external involvement has been in the form of commerce and finance (Reno, 1999).

Table 4.1.Outbreaks of Civil War, 1960–99
CountryStart of WarEnd of WarPrevious War
AngolaFeb. 1961Nov. 1975No
AngolaNov. 1975May 1991Yes
AngolaSept. 1992OngoingYes
BurundiApr. 1972Dec. 1973No
BurundiAug. 1988Aug. 1988Yes
BurundiNov. 1991OngoingYes
ChadMar. 1980Aug. 1988No
Congo, Republic of the1997Oct. 1997No
EthiopiaJuly 1974May 1991No
Guinea BissauDec. 1962Dec. 1974No
LiberiaDec. 1989Nov. 1991No
LiberiaOct. 1992Nov. 1996Yes
MozambiqueOct. 1964Nov. 1975No
MozambiqueJuly 1976Oct. 1992Yes
NigeriaJan. 1966Jan. 1970No
NigeriaDec. 1980Aug. 1984Yes
RwandaDec. 1963Feb. 1964No
RwandaOct. 1990July 1994Yes
Sierra LeoneMar. 1991Nov. 1996No
Sierra LeoneMay 1997July 1999Yes
SomaliaApr. 1982May 1988No
SomaliaMay 1988Dec. 1992Yes
SudanOct. 1963Feb. 1972No
SudanJuly 1983OngoingYes
UgandaMay 1966June 1966No
UgandaOct. 1980Apr. 1988Yes
Zaïre (Congo, Democratic
Republic of)July 1960Sep. 1965No
Zaïre (Congo, Democratic
Republic of)Sep. 1991Dec 1996Yes
Zaïre (Congo, Democratic
Republic of)Sep. 1997Sep. 1999Yes
ZimbabweDec. 1972Dec. 1979No

A number of factors have been identified as relevant in explaining civil wars in Africa. These factors include the historical legacies of slave trade and colonialism, the nature of the African state after independence, external intervention in the internal affairs of African countries, human rights abuses by African governments,11 and ethnic and religious grievances.12 The economic theory of civil war, however, sees civil wars in Africa as being caused by poor growth, poverty, and the abundance of natural resources such as diamonds, gold, and other precious minerals that can often be the source of much greed.

Section IV. The Model

The Collier-Hoeffler Model

The C-H model of civil war predicts the probability that a civil war will be initiated during a five-year period. The model is based on an analytic model that is in the rational choice tradition (Collier, 2000a). The model casts the causes of civil wars in terms of utility maximization. Rebels are rational individuals who will only initiate a rebellion if the expected benefit of the rebellion is greater than the cost. Mathematically, the C-H model can be written as follows:

where Y is the log of the odds ratio, or more specifically, the log odds of war.13 The variable i stands for the ith country and t for the tth time period. The explanatory variables are either measured at the beginning of the five-year period, or during the previous five-year period, or are time invariant. The χ variables are measured at the beginning of the period and include GDP per capita, primary commodity exports as a proportion of GDP, and population. The G variables are measured in the previous five-year period and include per capita income growth. The N variables change rather slowly or are time invariant and include social fractionalization.

Estimation Method

The analysis uses a panel data set of 44 sub-Saharan African countries for eight five-year periods (1960–64, 1965–69,…,1995–99) to estimate the probability that a large-scale civil war will be initiated in each five-year period using the logit regression approach. The logit is interpreted as follows: The slope coefficients measure the change in Y for a unit change in any of the explanatory variables, illustrating how the log odds change as the explanatory variables change by a unit. The predicted probability of war can be computed using the estimated coefficient for the above regression.

The probabilities for hypothetical observations can be calculated by first finding the average values for all explanatory variables for a subset of countries and taking this to represent a typical country within the subset. Then apply the following formula:

where e is the naperian log and y is the value of y [from equation (2)] using the estimated coefficients from the regression.

The Data

The dependent variable in this study is the risk of civil war (war start).14 The war start variable takes a value of one if a civil war started during the period and zero if the country is at peace. If a war started in period t and continues in t+1, the value of the war-start variable is recorded as missing. As mentioned previously, a civil war is defined as an internal conflict in which at least 1,000 battle-related deaths occurred. We use mainly the data collected by Singer and Small (1993) as in Collier and Hoeffler (2002a) and according to their definitions. The general source of the data (with the exception of the institutional variables) is Collier and Hoeffler (2002a). We also introduce additional variables such as institutional quality and corruption.

Definition of explanatory variables

  • Per capita income is the natural logarithm of per capita income. It is measured at the beginning of each subperiod.

  • Growth of per capita income is used as a proxy for economic opportunities. It is measured in the five-year period prior to the one for which the risk of civil war is being measured.

  • Primary commodity exports/GDP is used to proxy the abundance of natural resources. It is measured at the beginning of each subperiod.

  • (Primary commodity/GDP)2 is the square of the ratio of primary commodity exports to GDP and indicates high levels of primary commodity dependence.

  • Population is the natural logarithm of the population and it is measured at the beginning of the period. It is included to control for the size of a country’s population.

  • Social fractionalization is the combined measure of ethnic and religious fractionalization computed by Collier and Hoeffler.

  • Ethnic dominance is defined as occurring when the largest ethnic group constitutes between 45 and 90 percent of the population. It takes the value of one for ethnically dominant societies and zero otherwise.

  • Population dispersion measures the geographic dispersion of the population. A value of one indicates that the total population is concentrated in one area, and a value of zero indicates that the population is evenly distributed.

  • Peace duration measures the length of the period since the end of the previous civil war.

Measurement of institutional quality

Recent empirical analyses have typically considered three broad measures of institutions. The first is the quality of governance, including corruption, political rights, public sector efficiency, and regulatory burdens. The second is the legal protection of private property and law enforcement. The third is accountability and the limits placed on the executive and political leaders. All of these measures are subjective and are usually dependent on the perceptions and assessments of country experts or assessments made by residents responding to a survey (Edison, 2003).15

Using data from the International Country Risk Guide (PRS Group, 2003), we constructed an index of institutional quality. Because the International Country Risk Guide16 has data only for 32 countries in sub-Saharan Africa, we estimate the same regression for this subset of countries this time with the inclusion of the institutional variables. We take five institutional variables from the International Country Risk Guide to compute an index of institutional quality. The five variables are corruption, law and order, bureaucratic quality, democratic accountability, and government stability. We average across these five measures at the beginning of each subperiod to form the overall measure of institutional quality. Then, as in Barro (1991), we use the 1984 value for the 1980–84 period under the assumption that institutional quality changes slowly. We treat the 1960s and 1970s, however, as missing observations. The higher the value of the index, the better the quality of institutions in any given country.

Our measure of institutional quality has five components: law and order, bureaucratic quality, democratic accountability, government stability, and corruption. Law and order is an assessment of the strength and impartiality of the legal system as well as the popular observance of law. It ranges from zero to six. Bureaucratic quality measures the institutional strength and quality of bureaucracy. This measure is expected to be a shock absorber that minimizes reversions of policy when government changes. High points are given to countries where the bureaucracy is autonomous from political pressure and that have an established mechanism for recruitment and training. The variable ranges from zero to four.

Democratic accountability measures how responsive the government is to its people, on the basis that the less responsive it is, the more likely it is that the government will fall, peacefully in a democratic society, but possibly violently in a nondemocratic society. It ranges from zero to six. Government stability is an assessment of both the government’s ability to carry out its declared programs and its ability to stay in office. The components include government unity, legislative strength, and popular support. The variable ranges from 0 to 12. Corruption refers to corruption in the political system. The value ranges from zero to six. The higher the value of the corruption index the lower the level of corruption. In other words, countries that have a lower level of corruption have a higher value of the index and vice versa.

Section V. Regression Results

The Collier-Hoeffler Model

It is useful to begin by outlining the relevant factors that account for the initiation of civil war globally. Table 4.2, model 1 presents the results of the global model using a sample of 161 countries:17 GDP per capita is statistically significant and negatively related to the risk of civil war. Economic growth is statistically significant and negatively related to the risk of civil war. Primary commodity exports are significant and positively related to the risk of civil war. However, very high levels of primary commodity exports are negatively related to the risk of civil war. Therefore, the relationship between primary commodity exports and the risk of civil war is nonmonotonic or, more specifically, quadratic. The size of a country’s population is statistically significant and positively related to the risk of civil war. Ethnic dominance is positively related to the risk of civil war but statistically insignificant. Social fractionalization is significant and negatively related to the risk of civil war. That is, more fractionalized societies have lower risk of civil war. Peace duration is statistically significant and negatively related to the risk of conflict. That is, the longer the duration of the peace period, the lower the risk of civil war.

Table 4.2.The Baseline Collier-Hoeffler Results(Global sample)
Pooled Logit Estimates

Dependent Variable: Risk of Civil War

Variable
Model 1Model 2
Ln GDP per capita–1.0528–1.0529
(–3.64)***(–3.64)***
GDP per capita growth (t–1)–0.1027–0.1025
(–2.44)***(–2.44)***
Primary commodity exports/GDP16.69116.74
(3.23)***(3.22)***
(Primary commodity exports/GDP)2–23.532–23.35
(–2.36)**(–2.31)***
Peace duration–0.00373–0.0037
(–3.76)***(–3.74)***
Ln population0.4730.476
(3.45)***(3.38)***
Social fractionalization–0.00022–0.00022
(–2.10)**(–2.11)**
Geographic dispersion–0.994–0.992
(–1.10)(–1.09)
Ethnic dominance0.4490.450
(1.36)(1.36)
African dummy–0.371–0.3208
(–0.70)(–0.43)
Primary Africa–0.2678
(–0.09)
Pseudo R square0.2240.224
Log likelihood–146.60–146.60
Number of observations (N)750750
Source: IMF staff estimates.Notes: Ln is the natural logarithm. All regressions include a constant. T values are in parentheses. *** and ** indicate significance at the 1 and 5 percent levels, respectively. African dummy equals one if a country is in sub-Saharan Africa and zero otherwise. Primary Africa equals the interaction between the African dummy and primary commodity exports.
Source: IMF staff estimates.Notes: Ln is the natural logarithm. All regressions include a constant. T values are in parentheses. *** and ** indicate significance at the 1 and 5 percent levels, respectively. African dummy equals one if a country is in sub-Saharan Africa and zero otherwise. Primary Africa equals the interaction between the African dummy and primary commodity exports.

Is There an African Effect?

We would like to extend the C-H analysis to look at the specific case of sub-Saharan Africa, to investigate whether there is any hidden African effect. To do this, we estimate the C-H model using a panel data set of 44 countries in this region. We use pooled logit estimation techniques.

Comparative statistics

First it is important to examine some comparative statistics. Table 4.3 presents comparative statistics for a sample of sub-Saharan African countries during the sample period. The data show the following:

Table 4.3.Comparative Statistics
No Civil WarCivil War
MeanStandard

deviation
MinimumMaximumNMeanStandard

deviation
MinimumMaximumN
War0000302101130
Ln GDP per capita6.8340.6265.5498.8292916.4930.47075.4027.54028
Economic growth0.6953.803–9.90613.19252–1.843.606–10.592.88223
Primary exports/GDP0.1750.1390.090.5682850.17440.1250.0390.50530
(Primary exports/GDP)20.0500.0760.00080.3232850.0460.0660.00150.25530
Fractionalization3257.92062.42069753023847.62056.9180621030
Ethnic dominance0.40400.492013020.2330.4300130
Peace duration325.86162.261592302198.43166.50159230
Ln population14.7811.45610.63818.5329815.8511.04713.20318.08030
Geographic dispersion0.5460.21600.8582940.5670.1340.3080.80430
Institutional quality3.2030.8910.805.335922.31020.8470.7673.65328
Corruption2.7801.06606922.2771.2810428
Source: IMF staff estimates.Note: Ln is the natural logarithm.
Source: IMF staff estimates.Note: Ln is the natural logarithm.
  • Countries that experienced civil wars during the period had a lower GDP per capita than those countries that sustained peace.

  • Countries that experienced civil wars had a lower growth rate than countries that sustained peace.

  • Countries that experienced civil wars had on average about the same primary commodity exports–GDP ratio as those countries that sustained peace.

  • Countries that sustained peace had a much higher quality of institutions than those countries that had civil wars.

  • Sample countries that had civil wars were more corrupt than countries that sustained peace.

It is not surprising that the main finding from this comparative analysis is that countries that had civil war(s) have a lower endowment of growth-enhancing characteristics than those countries with no civil war(s).

Descriptive statistics

Table 4.4 presents the descriptive statistics for sub-Saharan Africa collectively. Three important points are worth mentioning: (1) between 1960 and 1999, the average risk of war for sub-Saharan Africa was about 9 percent; whereas (2) the average growth rate of per capita income for the region as a whole was about 0.42 percent for the sample period; and (3) during that period, about 17.2 percent of the region’s GDP was made up of the exports of primary commodities.

Table 4.4.Descriptive Statistics
VariableNMeanStandard

Deviation
MinimumMaximum
War start3320.09040.28701
Ln GDP per capita3456.7980.6325.4030.830
Growth3000.4193.735–10.4913.189
Primary exports/GDP3470.1720.1360.0090.568
(Primary exports/GDP)23470.0480.0740.000080.323
Fractionalization3603369.42022.2206975
Ethnic dominance3600.40.49101
Peace duration332314.35166.461592
Ln population35615.061.47010.63818.527
Geographic dispersion3520.5700.18900.858
Institutional quality1202.9950.9560.7575.335
Corruption1202.6621.13406
Source: IMF staff estimates.Note: Ln is the natural logarithm.
Source: IMF staff estimates.Note: Ln is the natural logarithm.

Correlations

At this point it is worth considering some important correlations (Table 4.5). For instance, GDP per capita is negatively correlated with the risk of civil war in Africa, as is economic growth. The size of a country’s population is positively correlated with the risk of conflict in Africa, as is its primary commodity exports. However, very high levels of primary commodity exports are negatively correlated with the risk of civil war. The corruption index is negatively correlated with the risk of war. Low corruption (high value of the index) is negatively associated with the risk of war. Our measure of institutional quality is negatively correlated with the risk of war.

Table 4.5.Correlations
VariableWarsaLnGDPGrowthSxpSxp2FracEtdo4590PeaceLnpopGeorgiaIndexCorrupt
Warsa1.000
LnGDP–0.1651.000
Growth–0.1790.2601.000
Sxp0.0040.4400.0491.000
Sxp2–0.0090.4300.0470.9651.000
Frac0.039–0.077–0.0350.1270.0601.000
Etdo4590–0.1130.2210.1780.0200.043–0.1521.000
Peace–0.2230.277–0.0320.1480.160–0.1590.2051.000
Lnpop0.226–0.454–0.223–0.172–0.2380.481–0.291–0.2641.000
Georgia0.014–0.19–0.1580.0140.0040.064–0.1610.0850.2001.000
Index–0.3530.4590.130–0.102–0.089–0.4180.2090.327–0.0090.1591.000
Corrupt–0.2500.3160.131–0.184–0.186–0.5400.2440.270–0.0230.0570.7261.000
Source: IMF staff estimates.Notes: Warsa is the risk of war; Ln is the natural logarithm; LnGDP is the natural log of GDP per capita; Growth is growth of GDP per capita; Sxp is primary commodity exports/GDP; Sxp2 is the square of primary commodity exports/GDP; Frac is social fractionalization; Etdo4590 is ethnic dominance; Peace is the duration of peace period; Lnpop is the natural log of population; Georgia is geographic dispersion of population; Index is institutional quality; and Corrupt is corruption.
Source: IMF staff estimates.Notes: Warsa is the risk of war; Ln is the natural logarithm; LnGDP is the natural log of GDP per capita; Growth is growth of GDP per capita; Sxp is primary commodity exports/GDP; Sxp2 is the square of primary commodity exports/GDP; Frac is social fractionalization; Etdo4590 is ethnic dominance; Peace is the duration of peace period; Lnpop is the natural log of population; Georgia is geographic dispersion of population; Index is institutional quality; and Corrupt is corruption.

Results

Table 4.6, model 1 presents the baseline results for 44 countries in sub-Saharan Africa:

Table 4.6.Regression Results for Sub-Saharan Africa
Pooled Logit Estimates

Dependent Variable: Risk of Civil War

Variable
Model 1Model 2Model 3
Ln GDP per capita–1.030727–1.144799–0.868315
(–1.80)*(–1.97)**(–1.72)*
GDP growth (t–1)–0.1119062–0.996007–0.1188602
(–1.66)*(–1.45)(–1.49)
Primary commodity exports/GDP6.711495.8867677.647955
(0.82)(0.71)(0.92)
(Primary commodity exports/GDP)2–4.697739–3.52968–5.723488
(–0.32)(–0.24)(–0.39)
Social fractionalization–0.0001796–0.0001996–0.0001953
(–1.36)(–1.46)(–1.47)
Peace duration–0.0031002–0.0030323–0.0028763
(–2.10)**(–2.04)**(–1.90)*
Ln population0.59003760.59144170.5760912
(2.28)**(2.28)**(2.16)**
Ethnic dominance–0.1260415–0.15002260.1002307
(–0.28)(–0.26)(0.17)
Geographic dispersion–0.4586954–0.6162363–0.2975275
(–0.28)(–0.36)(–0.19)
Diamond0.5912055
(1.04)
Great Lakes0.8807
(1.35)
N261261261
Pseudo R square0.19670.20340.2081
Log likelihood–62.517365–63.745447–61.629345
Source: IMF staff estimates.Notes: All regressions include a constant. T values are in parentheses; Asterisks (** and *) indicate significance at the 5 and 10 percent levels, respectively. Great Lakes is a dummy variable for countries in the Great Lakes region. It takes a value of one if a country is in the Great Lakes region and zero otherwise. Diamond is a dummy variable for diamond-exporting countries. It takes the value one if a country exports diamonds and zero otherwise. Ln is the natural logarithm while N is the size of the sample.
Source: IMF staff estimates.Notes: All regressions include a constant. T values are in parentheses; Asterisks (** and *) indicate significance at the 5 and 10 percent levels, respectively. Great Lakes is a dummy variable for countries in the Great Lakes region. It takes a value of one if a country is in the Great Lakes region and zero otherwise. Diamond is a dummy variable for diamond-exporting countries. It takes the value one if a country exports diamonds and zero otherwise. Ln is the natural logarithm while N is the size of the sample.
  • GDP per capita is significant and negatively related to the risk of conflict. Low GDP per capita increases the risk of civil war.

  • In addition, the rate of economic growth is negatively related to the risk of civil war. That is, as would be expected, higher rates of economic growth reduce the risk of civil war.

  • The size of a country’s population is statistically significant and positively related to the risk of civil war, so that countries with greater populations run a higher risk of civil war.

  • Peace duration, or the longer a country is at peace, is statistically significant and negatively related to the risk of civil war.

  • Social fractionalization is negatively related to the risk of civil war but not statistically significant.

  • Ethnic dominance is statistically significant but negatively related to the risk of civil war.

  • The dependence on a primary commodity, such as the DRC’s dependence on diamonds, is positively related to the risk of civil war. However, very high levels of exports for primary commodity exports are negatively related to the risk of civil war. Neither is statistically significant.

Initial per capita income and lagged per capita income growth are both statistically significant in explaining the probability of civil war in sub-Saharan Africa. Their level of significance improves when they are not both included in the same regression. One might wonder why the exports of a primary commodity are not significant in explaining the risk of civil war in Africa. First, compare the correlation between primary commodity exports in the global sample (0.0061) with the correlation between primary commodity exports and GDP in the African sample (0.4401), and one can see that the correlation between exports of a primary commodity and GDP in Africa is far higher than in the global sample.

Second, a cursory look at the comparative statistics reveals that those countries that experienced civil wars had on average almost the same primary commodity exports–GDP ratio as those countries that sustained peace. However, in the global sample, those countries that experienced civil wars had on average a much lower primary commodity exports–GDP ratio than countries that had experienced long periods of peace (15–17 percent). It is important to emphasize that other regions are also dependent on natural resources. However, since the relationship between natural resources and the risk of civil war is quadratic, what should be of interest is the standard deviation and not the mean (Elbadawi and Sambanis, 2000a). Our analysis shows that the standard deviation of African countries’ natural resource dependence is smaller than the standard deviation of non-African countries. Therefore, more African countries are closer to the peak of natural resource dependence, a factor that maximizes the threat of civil war. If this is the case, the coefficient of primary commodity dependency should be statistically significant. To examine this hypothesis we devise a test that can be seen in the next section.

Testing for the hypothesis of equal coefficients

We test whether the coefficient of primary commodity exports is statistically the same in and outside of Africa. A basic method in applied econometric research is to introduce a dummy variable for Africa in the regression of the global sample and then interact this dummy variable with primary commodity exports to form a new variable (call it primary Africa). We then estimate the model with both the African dummy variable and the primary Africa variable in the regression.18 A T-test on the coefficient of primary Africa is then conducted. Table 4.2, model 2 gives the results of this regression, showing that the coefficient of primary Africa is not statistically significant. Therefore, we fail to reject the null hypothesis that the coefficient of primary commodity exports in and outside of Africa is the same.

At this point, we test for an even stronger restriction19—an F-test on the null hypothesis that the coefficients of both the African dummy and the primary Africa variable are zero. This test examines the null hypothesis that not only is the coefficient of both the African dummy and the primary Africa variable the same in and outside of Africa, but also the probability of war is the same both in and outside of the continent when the full set of explanatory variables are equal. Thus, the test rejects this hypothesis.20 A possible explanation for why the coefficient of primary commodity exports is not statistically significant in Africa is the relatively high correlation between primary commodity exports and GDP in Africa. Therefore, in subsequent versions of the C-H model we omit the growth variables to examine this hypothesis.

Diamond-exporting countries

We also investigate those countries in sub-Saharan Africa that export diamonds. We investigate whether diamond-exporting countries21 in sub-Saharan Africa have a much higher risk of civil war.22 Fighting over diamond deposits is believed to have been an important reason for the initiation of civil wars, their maintenance, and prolongation in Angola, the DRC, Liberia, and Sierra Leone. The Fowler report (2000) on the role of diamonds in Angola argued that rough gems are important for the ability of the National Union for the Total Independence of Angola or UNITA (Union Nacional Por La Indepencee Totale do Angola) to sustain the rebellion. Diamonds, according to the report, allow the rebels to acquire new weapons, make friends, gain external support, and serve as a store of wealth. We evaluate this probability by introducing a dummy variable (call it diamond) for countries in the region that export diamonds. We then estimate the baseline regression this time with the inclusion of the diamond variable. Table 4.6, model 2 gives the result of this regression. The coefficient of diamond is positive, indicating that diamond-exporting countries have a higher risk of civil war but the coefficient is not statistically significant,23 meaning that countries that export diamonds do not behave differently from the rest of Africa.

Institutional Quality and the Risk of Civil War

We extend the C-H analysis by looking at the role of institutions in the initiation of a war by introducing institutional quality into the C-H model while dropping the growth variables. We measure institutional quality at the beginning of each five-year period. Data represent only 32 countries in sub-Saharan Africa. However, because our measure of institutional quality is highly correlated with GDP we exclude the growth variables from models 2 and 3. We use our measure of institutional quality to represent the strength and accountability of the state. Higher values of the index represent a higher opportunity cost of rebellion because of the strength of the state. Institutions with greater quality can also represent a lower level of criticisms because they offer good opportunities for resolving grievances. The result is presented in Table 4.7. In model 1, we have the results of the C-H model for 32 countries without any institutional variables. In model 2, we introduce institutional quality. The results show that the coefficient of our measure of institutional quality is highly significant (at the 1 percent level of significance) and negatively related to the risk of civil war. This implies that the development of good-quality institutions, such as rule of law, democratic accountability, efficient bureaucracy, and government stability, can reduce the risk of civil war in sub-Saharan Africa.

Table 4.7.Institutions and the Risk of Civil War
Pooled Logit Estimates

Dependent Variable: Risk of Civil War

Variable
Model 1Model 2Model 3
Ln GDP per capita–0.1725256
(–1.97)**
GDP growth (t–1)–1.1310005
(–1.65)*
Primary commodity exports/GDP15.0349137.9970329.80815
(1.42)(1.89)*(1.87)*
(Primary commodity exports/GDP)2–16.16715–55.00197–43.23131
(–0.91)(–1.67)*(–1.65)*
Social fractionalization–0.0001262–0.000329–0.0002666
(–0.63)(–1.02)(–0.87)
Peace duration–0.0042885–0.0077776–0.0081069
(–2.43)**(–2.50)**(–2.93)***
Ln population0.4227879–0.026684–0.2097021
(1.44)(–0.05)(–0.52)
Ethnic dominance0.76509910.57994710.0400845
(1.17)(0.56)(0.04)
Geographic dispersion2.8377449.5009867.399742
(1.05)(1.29)(1.57)
Institutional quality–1.497642
(–2.60)***
Corruption–0.7028002
(–1.70)*
N191104104
Pseudo R square0.23350.40890.3340
Log likelihood–45.712473–23.1613–26.09777
Source: IMF staff estimates.Notes: All regressions include a constant. T values are in parentheses. Asterisks (***, **, and *) indicate significance at the 1, 5, and 10 percent level, respectively. N is the size of the sample while Ln is the natural logarithm.
Source: IMF staff estimates.Notes: All regressions include a constant. T values are in parentheses. Asterisks (***, **, and *) indicate significance at the 1, 5, and 10 percent level, respectively. N is the size of the sample while Ln is the natural logarithm.

In this analysis, primary commodity exports, this time, are statistically significant and have the expected positive sign. The square of primary commodity exports is also significant and has the expected negative sign. Peace duration is found to be significant with the expected negative sign, and social fractionalization has the expected negative sign but is not statistically significant. Ethnic dominance has the expected positive sign but is not statistically significant. The size of a country’s population takes on a negative sign but is not statistically significant. Geographic dispersion of population is statistically insignificant and has an unexpected wrong sign.

In Table 4.7, model 3, we introduce corruption into the C-H framework. This measure of corruption, a component of the measure of institutional quality, is from the International Country Risk Guide. Again, because of the apparent correlation with GDP we exclude GDP from the regression. Corruption is significant at the 10 percent level of risk and has the expected negative sign. The interpretation of this variable is this: higher levels of the index (low corruption) reduce the risk of civil war. Peace duration is again statistically significant with the expected sign. Social fractionalization has the expected negative sign but is statistically not significant. Ethnic dominance has the expected positive sign but is statistically not significant. The size of the population is not statistically significant and has a negative sign. The geographic dispersion of the population is not statistically significant and has the wrong sign.

The introduction of the institutional variables (institutional quality and corruption) and the removal of the growth variables make primary commodity exports and its square significant at the 10 percent level of risk. This confirms the quadratic relationship between primary commodity exports and the risk of civil war. Primary commodity exports increase the risk of civil war, but very high levels of primary commodity exports reduce the risk of civil war. The result of this regression to some extent confirms our hypothesis that the insignificance of primary commodity exports in the baseline model is due to the strength of the relationship between primary commodity exports and GDP. The correlation between primary commodity exports and GDP is 0.44, but that of primary commodity exports and institutional quality and corruption is –0.1015 and –0.1839, respectively. Thus, there is indeed an African effect; however, this effect does not manifest itself through the outright rejection of the C-H model, but reflects in the coefficients of some important variables as we have seen in the case of primary commodity exports and GDP.

Estimation Issues: The Random-Effects Probit Model

We investigate whether the results of the chapter are affected by the choice of estimation techniques, using panel data techniques to investigate this hypothesis. Given the panel nature of our data set, the choice of any estimation technique will depend on the assumptions about the distribution of the error term (µit) and assumptions about the structure of the cross-section effects. If the µit’s are taken to be independent standard normal variables, then the panel nature of the data is irrelevant, and the pooled logit estimation method is the most appropriate (Greene, 1997). If not, then the random-effect probit estimator is more efficient because it gives us the flexibility to model cross-sectional differences that are not possible to explore using the pooled logit. The error term in a random-effects model can be specified as

where vi is the individual specific effects. Both components are normally distributed with mean zero and independent of one another.24 We assume that vi are random because some of our explanatory variables such as social fractionalization are either time invariant or changing rather slowly. The random-effects probit model makes it possible to capture some of the unobserved heterogeneity among the cross sections, and if these effects are significant then the model is more efficient than the pooled logit.25 The results from the random-effects probit are presented in Table 4.8. In model 1, we have the results for the baseline model (44 countries). In model 2, we include institutional quality for 32 countries. These results are statistically similar to those reached using the pooled logit. In fact, a likelihood ratio test of the correlation coefficient (Rho) shows that the panel variance is not significant in explaining total variance; it is correct to use the pooled logit model.26 Therefore, we can conclude that our previous result is not based on the choice of the estimation method.27

Table 4.8.Random-Effects Probit Regression
Dependent Variable: Risk of Civil War

Variable
Model 1Model 2
Growth of GDP per capita (t–1)–0.0685773
(–1.86)*
GDP per capita–0.4982207
(–1.72)*
Primary commodity exports/GDP2.76513519.27667
(0.68)(1.80)*
(Primary commodity exports/GDP)2–1.098504–27.56934
(–0.15)(1.55)
Social fractionalization–0.0000799–0.0001856
(–1.19)(–1.15)
Ethnic dominance–0.05175990.2779416
(–0.18)(0.49)
Ln population0.3139633–0.0019025
(2.33)**(–0.01)
Peace duration–0.0015713–0.0039572
(–2.04)**(–2.47)**
Geographic dispersion of population–0.23896425.037007
(–0.29)(1.09)
Institutional quality–0.8227104
(–2.59)***
Rho8.32e-078.32e-07
No. of observations261104
Log likelihood–62.485802–23.332053
Source: IMF staff estimates.Notes: T values are in parentheses. Asterisks (***, **, *) indicate significance at the 1, 5, and 10 percent level, respectively. Ln is natural logarithm.
Source: IMF staff estimates.Notes: T values are in parentheses. Asterisks (***, **, *) indicate significance at the 1, 5, and 10 percent level, respectively. Ln is natural logarithm.

Performance of the C-H Model

The C-H model predicts the risk of civil war for sub-Saharan Africa to be 8.8 percent for the period 1960–99. This prediction is done at the means of the explanatory variables. The actual risk of civil war for Africa during the sample period is 9 percent, which is close to the predicted probability. Therefore, the C-H model predicts the risk of civil war in sub-Saharan Africa with considerable accuracy (see Figure 4.1). However, for individual countries, the C-H model systematically underpredicts the probability of war for high-risk countries and overpredicts the probability of civil war for low-risk countries (see Table 4.9).

Figure 4.1.Probability of Civil War in Sub-Saharan Africa

Source: IMF staff estimates.

Table 4.9.Synthetic Index of the Risk of Civil War in Sub-Saharan Africa, 1960–99
Low-Risk

Countries
Actual

%
Predicted

%
Moderate-Risk

Countries
Actual

%
Predicted

%
High-Risk

Countries
Actual

%
Predicted

%
Benin01.84Congo, Rep. of12.291.8Angola100.0057.85
Botswana00.6Gabon08.4Burundi42.8523.48
Burkina Faso05.8Guinea07.3Chad14.299.64
Cameroon06.3Kenya08.0Ethiopia25.0028.74
Cape Verde01.1Madagascar011.11Guinea Bissau16.675.00
Central Africa Rep.102.25Malawi08.2Liberia28.576.40
Comoros05.7Mali08.2Nigeria46.2933.19
Djibouti00.54Mauritania08.6Rwanda46.2921.19
Gambia03.4Niger08.6Sierra Leone46.296.40
Ghana05.6Tanzania08.4Somalia48.8016.67
Côte d’Ivoire06.0Zambia010.19Sudan66.6714.68
Lesotho01.9Uganda28.5716.34
Namibia00.6Zaïre (Congo,
Senegal04.8Dem. Rep. of)42.8534.41
Seychelles00.0Zimbabwe14.2911.21
South Africa03.6
Swaziland01.9
Togo04.4
Source: IMF staff estimates.Notes: Predicted probability of civil war is calculated using the coefficients from the baseline regression and the mean values for the explanatory variables for every country and then applying the formula in equation (2). We classify a country as having low risk of civil war if the predicted and/or actual probability of civil war is less than 7 percent. If a country has between 7 and 14 percent probability of war, we classify the country as having a moderate risk of war. We classify a country to be in the high-risk category if the probability of civil war is greater than 14 percent.

Even though the Central African Republic has seen some violent conflicts, these conflicts do not qualify as civil wars by the Singer-Small definition. This, among other factors, may account for its low risk of civil war.

Source: IMF staff estimates.Notes: Predicted probability of civil war is calculated using the coefficients from the baseline regression and the mean values for the explanatory variables for every country and then applying the formula in equation (2). We classify a country as having low risk of civil war if the predicted and/or actual probability of civil war is less than 7 percent. If a country has between 7 and 14 percent probability of war, we classify the country as having a moderate risk of war. We classify a country to be in the high-risk category if the probability of civil war is greater than 14 percent.

Even though the Central African Republic has seen some violent conflicts, these conflicts do not qualify as civil wars by the Singer-Small definition. This, among other factors, may account for its low risk of civil war.

So What Explains Civil Wars in Sub-Saharan Africa?

The empirical analysis performed leads us to conclude that civil wars in Africa have both economic and political undertones (see Box 4.1). Economically, Africa’s high level of poverty has been a crucial causal factor in most of its civil wars. Collier and Hoeffler (2002a) interpreted low GDP per capita to mean a much lower opportunity cost of rebellion, encouraging rebels to launch an attack. Fearon and Laitin (2000), however, use GDP per capita to proxy the strength of the state. It is important to emphasize that these two explanations are essentially the same. Higher GDP per capita implies higher strength of the state, which acts as a disincentive to rebels because they are less apt to challenge the might of the state. No matter how GDP per capita is interpreted, one important issue still remains: high GDP per capita reduces the risk of civil war.

Another important factor that explains the risk of civil war in sub-Saharan Africa is the rate of growth of GDP per capita. Africa’s relatively low growth rate has been an important determinant of the strings of war on the continent. Collier and Hoeffler interpreted faster economic growth as indicating the difficulty of rebellion movements in recruiting workforce. Faster economic growth represents better life opportunities for potential rebel recruits and also lower levels of grievances. The abundance of primary commodities is also associated with higher risk of conflict. However, extremely abundant natural resources reduce the risk of civil war. Collier and Hoeffler define a threshold level of about 26 percent. However, very few countries are at this level. While the coefficient of primary commodity exports in Africa is not statistically significant, there is no evidence that the coefficient of primary commodity exports is statistically different in and outside of Africa.

This finding leads us to conclude that the insignificance of primary commodity exports is the result of the strength of the relationship between primary commodity exports and GDP per capita in Africa.28 Collier and Hoeffler interpreted the abundance of primary commodities as offering an opportunity for rebels to finance themselves and thereby increase the risk of civil war. However, extremely plentiful natural resources leave the government with enough funds to defend itself. The primary commodity result should be subjected to further empirical investigations. A lot of work has to be done on the industrial organization of rebel movements before we can sufficiently say that diamonds, or other natural resources, are driving civil wars. As Herbst (2000) points out, most of the regressions reporting a relationship between resources that may be stolen and conflict may simply be picking up the fact that these resources may be necessary for conflict to continue but are not the driving force of conflict.

Box 4.1.Factors Behind the Risk of Civil War in Sub-Saharan Africa

  • Africa’s high level of poverty,

  • Africa’s relatively low growth of per capita income,

  • the abundance of primary commodities,

  • peace duration—the longer a country is able to prolong the period of peace, the lower the risk of civil war,

  • large populations,

  • Africa’s difficulty in developing strong institutions, and

  • the high level of corruption.

We find peace duration to be a significant factor in explaining the probability of civil war in sub-Saharan Africa. The longer the duration of the peace period, the lower the risk of civil war and vice versa. This confirms the assertion that time is an important factor in the healing process. Population size is also important in explaining the risk of civil war in sub-Saharan Africa. The larger the population, the higher the risk of civil war and vice versa. In other words, civil wars seem more likely in countries with large populations. The theoretical argument that links population to the risk of civil war is that the larger the population, the easier it should be to find a group to challenge the government.

Moreover, Africa’s difficulty in developing strong institutions can be blamed for the high incidence of civil war on the continent. As this empirical analysis shows, good-quality institutions such as rule of law, democratic accountability, government stability, efficient bureaucracy, and limited corruption are associated with a lower risk of civil war. A similar result was found by Easterly (2000),29 who uses a measure of economic governance and shows that good-quality institutions can mitigate the negative effects of ethnic diversity on economic growth. Rodrik (1999) finds high quality of economic or political institutions to reduce the effect of ethnic diversity on persistence of economic growth following external shocks. The evidence from our analysis is consistent with the evidence from the empirical literature. Good-quality institutions can support peaceful ways of resolving disputes and thereby reduce any adverse effects of ethnic dominance and social fractionalization. Indeed, countries with good-quality institutions have a lower risk of conflict.

There is no significant relationship between social fractionalization and the risk of civil war. It is important to stress that the coefficient of social fractionalization is negative, indicating that more fractionalized societies have a lower risk of civil war. This evidence provides some hope for sub-Saharan Africa whose ethnically diverse societies may have been blamed for the region’s string of civil wars. Africa’s ethnic diversity actually promotes peace and makes the region relatively safe. Collier and Hoeffler interpreted higher level of social fractionalization as representing the difficulty of coordinating a rebellion.

The Case of the Great Lakes Region

The Great Lakes region has witnessed some of Africa’s most intense and violent conflicts. Almost all the countries in this region have seen some period or periods of violent conflict.30 Among these violent conflicts are the genocides in Rwanda and Burundi and the regionalization of conflict in the DRC. Various conflicts that have affected the subregion have resulted in a large number of victims who have been displaced or forced to live in looted and destroyed societies. In a region in which most of the people have almost always lived very close to or below the poverty line, the continued deterioration of social, economic, and political life has resulted in the persistent uprooting and marginalization of fragile and vulnerable populations (United Nations, 1999). This empirical analysis is extended to cover countries in the Great Lakes region.

First, we compare the descriptive statistics for the Great Lakes region with that of sub-Saharan Africa. Table 4.10 presents the descriptive statistics for this subregion:

Table 4.10.Descriptive Statistics, Great Lakes Region
VariableMeanStandard

Deviation
MinimumMaximumN
War start0.270.4550129
Growth of per capita–0.583.982–10.487.1328
Ln GDP per capita6.330.3005.4026.8132
Ln population16.270.75414.8917.5932
Peace duration165.44119.76946329
Primary exports/GDP0.110.0440.0520.27332
(Primary exports/GDP)20.0140.0130.00270.07432
Social fractionalization4490.72515180621032
Ethnic dominance000032
Geographic dispersion0.560.1840.2910.80932
Institutional quality2.731.0510.84.0316
Corruption1.981.3440416
Source: IMF staff estimates.
Source: IMF staff estimates.
  • The average risk of war for the Great Lakes region for the period from 1960 to 1999 is 27 percent. This is far higher than the risk of war for the entire continent of sub-Saharan Africa (9 percent).

  • The growth of per capita income for the Great Lakes region between 1960 and 1999 was –0.58. During that period, the average growth rate for sub-Saharan Africa was 0.42 and the average for countries in Africa that experienced civil wars was –1.84.

  • The average GDP per capita for the Great Lakes region is lower than the average for sub-Saharan Africa. It is also lower than the mean of countries in conflict.

  • The region has a primary commodity exports–GDP ratio of 11 percent. This is lower than the average for sub-Saharan Africa (17 percent).

  • The Great Lakes region has a much more fractionalized society than sub-Saharan Africa as a whole.

  • Countries in the Great Lakes region have a much lower index of institutional quality than sub-Saharan Africa.

  • Countries in the Great Lakes region have a much lower index of corruption and are therefore more corrupt than sub-Saharan Africa as a whole.

Second, we investigate whether the factors that explain conflict in Africa can be used to explain the higher incidence of civil war in the Great Lakes region. To do this, we introduce a dummy variable into the baseline regression for Africa. The dummy takes the value of one if a country is in the Great Lakes region and zero otherwise. We estimate the C-H model with the inclusion of the Great Lakes dummy. The result is presented in Table 4.6, model 3. Clearly, the coefficient of the Great Lakes dummy is positive, indicating that countries in the Great Lakes region have a higher probability of civil war, but this is statistically not significant. This leads us to conclude that the probability of conflict in the region could be explained by the same factors that account for civil wars in sub-Saharan Africa (economic growth, per capita income, primary commodity exports, institutional quality, and corruption). Whereas the subregion’s relatively high level of social fractionalization makes this region less vulnerable to conflict, its relatively low level of economic growth and lower initial GDP makes the region more prone to civil war.

Third, we investigate the extent and accuracy to which the C-H model predicts the risk of war in this region. Figure 4.2 presents both the actual and the predicted risk of war for the Great Lakes region from 1965 to 1999. As evident from the diagram the C-H model systematically underpredicts31 the risk of war for the region. However, for the period 1995–99 the C-H model predicted perfectly the risk of war for this region to be 33 percent (exactly the same as the actual risk of war). The average risk of war for the Great Lakes region during the sample period (1965–99) is 27 percent. The C-H model predicted the average risk of civil war for the region to be 20 percent.

Figure 4.2.Probability of Civil War in the Great Lakes Region

Source: IMF staff estimates.

Section VI. Policies for Conflict Prevention and Peace Building

A peaceful and stable Africa is the wish of African countries as well as the international community. Before assessing the different policies that might help achieve permanent peace, it is important to understand the issues involved. As stated at the beginning of this chapter, conflict is complex. It does not take place just because there is greed or dissatisfaction. Neither does it occur because diamonds are abundant or because the government and social political institutions are weak. It occurs when these factors at multiple levels come together and reinforce one another. It is the result of the interaction of weak political and social institutions, deep grievances, poverty, bad governance, and global and regional geopolitics. Although most African countries are anxious for peace, they do not seem to know how to maintain it when they achieve it. As Bigombe and others (2000) point out, “civil wars always end but they usually restart,” especially in Africa. The strategy then should focus on conflict prevention and the maintenance of permanent peace.32

For postconflict countries, the best strategy should be to prevent a new war (see Box 4.2). As this empirical analysis shows, the longer the peace period lasts, the lower the risk of further conflicts. The end of a conflict does not suggest that the problems are over. The countries involved will face massive and onerous challenges of postwar reconstruction. Strangely enough, it is during this period that international attention shifts to other areas and regions, thereby losing valuable time in preventing a new war (Wallensteen, 2002). Thus, in a postconflict situation the priority should be to support the peace agreements that exist and to prevent the tension from reoccurring. Therefore, all planning for global development assistance should focus on observing the dangers of future war and finding means of mitigating such dangers. It is also important to entrench postconflict peace building into global development policies. Key development projects, such as the Millennium Development Goals, should clearly and explicitly focus on reducing the risk of conflict. As a country continues to build a permanent peace, the conflict is forgotten, new concerns emerge, and the question of systematic conflict prevention becomes paramount. It is here that policies of good governance, economic growth, poverty reduction, and institutional development come in. Here, too, it is imperative to stress that every conflict situation is different. Knowing the structure of the risk factors in a specific country should provide some guidance to policy prioritization on which factors to target for policy action (Bigombe and others, 2000).

As our empirical evidence shows, some of the factors that are crucial in African conflicts are dependence on natural resources, poverty, slow growth, weak institutions, and corruption. All of these are amenable to policy alteration. It is important to emphasize that related measures are not mutually exclusive, and a good combination of them should achieve the best results. Therefore, policies for building permanent peace should be a multidimensional package. It is important to design and implement a holistic approach to conflict prevention that encompasses policies aimed at reducing all important risk factors.

Box 4.2.Policy Implications

  • For postconflict countries, the best strategy should be to prevent new war. The main priority should be to support the peace agreements that exist and to prevent the tension from reoccurring.

  • It is important to entrench postconflict peace building into global development policies. Key development projects such as the Millennium Development Goals should clearly and explicitly focus on reducing the risk of conflict.

  • Postconflict economic policies should, therefore, focus on social reconciliation and reconstruction and tackle macroeconomic imbalances.

  • Policies to promote growth and reduce poverty will help in conflict prevention. The greatest gain in conflict prevention will be made in focusing development efforts on very poor countries.

  • Reducing primary commodity dependency should help prevent civil wars. In the short term, the international community should take appropriate measures to end the use of primary commodities that can be easily looted by rebel movements. This, however, will require a strong commitment from the international community.

  • In the long term, African governments should take positive steps to ensure rapid diversification of their economies. Growth, aid, and policy should help in this direction. The role of international financial institutions in the development and implementation of good and adequate economic policies becomes paramount.

  • An important approach to conflict prevention is the development of good-quality institutions such as rule of law, democratic accountability, efficient bureaucracy, government stability, and low corruption. Evidence that good-quality institutions become effective when a country reaches the middle-income level supports this policy option.

The econometric evidence presented in this chapter suggests that increasing the level and the growth of per capita income should reduce the risk of civil war. Faster economic growth and lower poverty present young people with greater hopes for the future, although policies to promote growth are beyond the scope of this chapter.33 Poverty is a factor considered by political and economic theorists as favoring rebellions. Poverty reduction increases the opportunity cost of rebellion because the poor have little to lose by joining a rebellion. At the same time, poverty reduction reduces grievances against the government. The core of the approach to poverty reduction should be to reduce unemployment, because it serves to reduce the pool of potential recruits for civil war. Therefore, better employment opportunities may be an effective medium-term strategy for conflict prevention. It is important to emphasize, however, that given Africa’s initial conditions, this route to conflict prevention may be a slow process.

Natural resource dependence is an important source of civil wars in sub-Saharan Africa. Primary commodity exports can increase the risk of conflict in four ways: financing rebels, worsening corruption in government, increasing the incentive for secession, and increasing exposure to shocks (Collier and others, 2003). The relationship between primary commodity exports is quadratic, implying that many African countries may reduce risk by increasing their resource dependence. However, because most African countries have primary resource dependence lower than the peak, the best policy will be to reduce the dependence on primary commodity exports. In the short term, the international community should take measures to end the use of natural resources that can be easily looted by rebel movements. This, however, will require a strong commitment on the part of the international community. In the long term, African governments should take positive steps to ensure rapid diversification of their economies. Collier and Hoeffler (2000b) find three measures that can help reduce the dependence on primary commodity exports: growth, aid, and policy. The implementation of good and adequate economic policies is of paramount interest because it advocates a role for the International Monetary Fund and the World Bank in conflict prevention.34

As the comparative statistics show, countries with high-quality institutions have a lower risk of civil war. An important approach for conflict prevention is the development of good-quality institutions such as rule of law, democratic accountability, efficient bureaucracy, government stability, and low corruption. In the case of sub-Saharan Africa, this calls for the reform and development of high-quality political and governance institutions. If the key to conflict prevention is the ability of the entire population to raise issues and address them in nonviolent ways through local organizations and community groups, then the establishment of formal political institutions becomes paramount. Guidelines for getting involved in political action and how authorities relate to them become critically important—something many analysts consider as the main feature of democracy (Wallensteen, 2002). Democracy functions best as a conflict prevention strategy when the stakes of the political contests are low (Weingast, 1997). The development of political, legal, and economic institutions that help reduce the risk of war by reducing the gains to narrow interests are critical for a well-functioning democracy. Evidence that political institutions are effective once a country reaches middle-income levels strongly supports this policy option. It is equally important to develop a good-quality bureaucracy. The bureaucracy should be independent from government manipulation, and recruitment should be open to all groups and carried out in full transparency. Also, African countries should take strong measures to reduce corruption in government. The development of good-quality institutions with good checks and balances should help in this direction.

Finally, it is important to remind ourselves that conflict prevention and peace building usually require huge financial and human resources, and that African governments do not have enough resources to actually bear these costs themselves. This calls for an urgent, purposeful, and effective role for the international community (including international financial institutions) not only to end civil wars but also to ensure the sustainability of durable peace. It is important to emphasize that the performance of the international community in this direction has so far been below the expectations of African countries. This perception, according to Annan (1998, s.11), has left a “poisonous legacy that continues to undermine confidence.”

Appendix 4.1. Data Definitions and Sources

The main source of data for almost all the variables that appear in this chapter is Collier and Hoeffler (2002a). The definitions of the variables are given below.

Per capita income. The variable, per capita income, is the natural logarithm of per capita income. Per capita income is measured as real purchasing power parity adjusted GDP per capita to provide reasonable comparability across countries. The primary data set is the Penn World Table 5.6 (Summers and Heston, 1991). Income data are measured at the beginning of each subperiod, 1965, 1970, …, 1995.

Growth rate of per capita income. Using the above income per capita measure, Collier and Hoeffler calculated the average annual growth rate as a proxy for economic opportunities. This variable is measured in the five-year period prior to the one for which conflict risk is being measured.

Primary commodity exports/GDP. This is the ratio of primary commodity exports to GDP and proxies the abundance of natural resources. The data on primary commodity exports as well as GDP were obtained from the World Bank. Export and GDP data are measured in current U.S. dollars. The data are measured at the beginning of each subperiod, 1965, 1970, … 1995.

Population. This is the natural logarithm of the population. The data source is the World Bank’s World Development Indicators 1998. Again, population is measured at the beginning of each subperiod—included to control for the size of the country’s population. If the effect of population were neutral, then one would expect that a doubling of population would approximately double the risk of conflict. For instance, this would be the assumption if two identical neighboring countries are merged. Empirically peaceful countries have, on average, less than half the population of conflict countries, higher population density, and are more urbanized.

Social fractionalization. If intergroup hatred is a crucial factor in civil wars, then it might be expected that homogeneous societies would be considerably safer than fractionalized societies. Social fractionalization is proxied using the Collier and Hoeffler combined measure of ethnic and religious fractionalization. Ethnic fractionalization is measured by the ethno-linguistic fractionalization index. It measures the probability that two randomly drawn individuals from a given country do not speak the same language. Data are only available for 1960. Using data from Barrett (1982) on religious affiliations, Collier and Hoeffler constructed an analogous religious fractionalization index. Following Barro (1991) they aggregated the various religious affiliations into nine categories: Catholic, Protestant, Muslim, Jew, Hindu, Buddhist, eastern religions (other than Buddhist), indigenous religions, and no religious affiliation. The fractionalization indices range from 0 to 100. A value of 0 indicates that the society is completely homogeneous, whereas a value of 100 would characterize a completely heterogeneous society. Collier and Hoeffler calculated the social fractionalization index as the product of the ethno-linguistic fractionalization and the religious fractionalization index plus the ethno-linguistic or the religious fractionalization index, whichever is greater.

Ethnic dominance. Collier and Hoeffler define this as occurring where the largest ethnic group constitutes 45-90 percent of the population. Using the ethno-linguistic data from the original data source (Atlas Narodov Mira, 1964) Collier and Hoeffler calculated an indicator of ethnic dominance. This variable takes the value of one if one single ethno-linguistic group makes up 45 to 90 percent of the total population and zero otherwise. According to Collier and Hoeffler, societies characterized by ethnic dominance have about double the risk of conflict of other societies.

Population dispersion. This variable measures the geographic dispersion of the population. Collier and Hoeffler constructed a dispersion index of the population on a country-by-country basis. Based on population data for 400-square-kilometer cells, they generated a Gini coefficient of population dispersion for each country. A value of zero indicates that the population is evenly distributed across the country, and a value of one indicates that the total population is concentrated in one area. Data are available for 1990 and 1995. For years prior to 1990, we use the 1990 data.

Peace duration. This variable measures the length of the peace period since the end of the previous civil war. For countries that never experienced a civil war we measure the peace period since the end of World War II until 1962 (172 months) and add 60 peace months in each consecutive five-year period.

Bibliography

For an excellent analysis of the relationship between civil wars and economic growth in sub-Saharan Africa, see Gyimah-Brempong and Corley (2001).

See Foreword of Stein (2003).

At the aggregate level of analysis the effect of civil wars on economic development includes the destruction of both human and nonhuman capital, the disruption of economic transactions, the distortion of the decision-making process by economic agents, and the disruption of efficient resource allocation. It has been estimated that during a war period the per capita growth rate of a country declines by 2.2 percentage points compared with the normal situation. See Collier (1999), for example.

Using the Collier and Hoeffler (2002a) data set, the values for the period 1960—99 are $4,219 for no-war countries and $1,645 for war countries. The average growth rate of per capita income for war countries is 0.226 percent and that of no-war countries is 1.74 percent.

Two main econometric studies have looked at the case of Africa: Collier and Hoeffler (2000b) and Elbadawi and Sambanis (2000b). Collier and Hoeffler introduced a dummy variable for Africa in their global model and concluded that there is no mysterious African effect. Elbadawi and Sambanis drew inferences from their global model and concluded that conflicts in Africa can be explained by the same global factors.

North (1994) defines institutions as the humanly devised constraints that structure human interaction. Bardhan (1989) also defines institutions as the “social rules, conventions and other elements of the structural framework of social interactions.” These institutions can be informal, such as social capital and norms, or formal legal rules such as laws ensuring individual liberties, property rights, and enforcement of contracts.

Ragnar Gudmundsson presents a detailed overview of the literature on alleviating poverty in postconflict countries elsewhere in Chapter 3.

The main slogan in the American revolution was that “all men are created equal.” In the French revolution, the slogan was “liberty, equality, fraternity.” The motto of the Russian revolution was “peace, land, bread,” and the Chinese communist revolution had the slogan “those who have much give much, those who have little give little.”

However, Elbadawi and Sambanis (200b) examine the case of Africa by drawing inferences from a global sample. As previously mentioned, this approach ignores vital regional and historical differences. In contrast, this chapter builds on the application of a variant of the C-H model to the African data set per se.

The Organization of African Unity, now African Union (AU), identifies human rights violations, ethnic rivalries, and clan and other factional rivalries as relevant causal factors in explaining the incidence of civil war in Africa.

Many countries in Africa are diverse ethnically, with more than 100 ethnic and language groups in Nigeria and the Democratic Republic of the Congo. Other countries, such as Burundi and Rwanda, are ethnically polarized with the Hutus comprising about 85 percent of the population and the Tutsis the remaining 15 percent.

The odds ratio in favor of war outbreak is the ratio of the probability of a war outbreak to the probability of no war outbreak in any given five-year period. Mathematically, the odds ratio is written as p/1–p.

Detailed explanation of all explanatory variables as given by Collier and Hoeffler (2002a) is presented in Appendix 4.1.

The aggregate governance index is the average of six measures of institutions developed in 1999 by Kaufman, Kraay, and Zoido-Lobaton (1999). The major components are voice and accountability, political stability and absence of violence, government effectiveness, regulatory burden, and freedom from graft. Other measures of institutions focus on property rights and constraints on the executive. Our measure of institutional quality is quite similar to that of Kaufman, Kraay, and Zoido-Lobaton (1999).

The definition of the component of the institutional index is from the International Country Risk Guide. See the International Country Risk Guide for a detailed explanation and the computations of the various components of our index of institutional quality.

Using Collier and Hoeffler’s data set and a pooled logit estimation technique, we reestimated their model and generated the results shown in Table 4.2, model 1. The result is similar to that of Collier and Hoeffler.

We include both the African dummy and primary Africa variable in the regression to be sure that any rejection of the null hypothesis would not be due to the fact that the African dummy is excluded.

We thank an anonymous referee for suggestions on the construction of these tests.

The chi-square value is 0.06 and the p value is 0.8079, thus making it possible to reject the null hypothesis at the 5 percent level of significance.

The major diamond-exporting countries in sub-Saharan Africa are Angola, Botswana, Central African Republic, Congo, Cote d’Ivoire, DRC, Ghana, Guinea, Liberia, Namibia, Sierra Leone, South Africa, and Zimbabwe.

This issue is investigated because of the supposed fatal role diamonds are believed to have played in several conflicts in Africa. Because of this, the United Nations has imposed sanctions on conflict diamonds originating from areas controlled by illegitimate rebel groups.

If we exclude primary commodity exports and its square from the baseline regression, the significance of the diamond variable improves marginally (say, significant at a 20 percent level of risk).

Full details on the estimation and inferences may be found in Butler and Moffitt (1982) and Greene (1995).

Because some of our variables are time invariant, the use of a fixed effect would create perfect multicollinearity between the nontime varying explanatory variables and the individual specific effects, forcing us to drop them from the regression. Furthermore, the logit does not lend itself well to random-effects treatment. See Greene (1997) for details.

The null hypothesis is that Rho is equal to zero. The chibar2 (01) = 0.00 and the P value is 1.00 in both models 1 and 2. We therefore do not reject the null hypothesis.

Collier and Hoeffler (2002a) further estimated their baseline regression using King and Zeng’s (2001) rare events correction procedure. They find the difference between the rare events logit and the standard logit to be negligible.

Using primary commodity exports as a proportion of GDP to capture the abundance of natural resources may be misleading in the African context. This ratio can also represent trade openness (if we use total exports). Because in most African countries primary commodities constitute about 60 percent of total exports, the ratio may be capturing openness rather than the abundance of natural resources.

Easterly (2000) constructed an index of institutional quality, which is the average of Knack and Keefer’s 1995 measures from the International Country Risk Guide of freedom from government repudiation of contracts, freedom from expropriation, rule of law, and bureaucratic quality.

According to Relief Web, the countries in the Great Lakes region are Burundi, the DRC, Kenya, Rwanda, Tanzania, and Uganda. Although the United Nations excludes Kenya from the Great Lakes region, this exclusion from the analysis does not significantly change the results for the Great Lakes region.

This may be because the sample size is small—consisting only of the six countries in the Great Lakes region. In larger samples, the model’s predictions improve.

Globally, within the first decade of the end of a conflict, 31 percent of them have resumed. In Africa, about half of peace restorations have lasted fewer than 10 years. See Bigombe and others (2000) for details.

See Chapters 6 and 7.

Collier and others (2003) measure the effectiveness of policy using the World Bank’s Country Policy and Institutional Assessment (CPIA). They find that an improvement of one point in the CPIA (which is approximately equivalent to the difference between African and South Asian policies) would reduce the dependence on primary commodities from 15.2 percent of GDP to 13.8 percent.

    Other Resources Citing This Publication