September 13, 2006 Democracy Out of Anarchy: How Do Features of A Civil War Influence the Likelihood of Post-Civil War Democracy?

Transcription

1 September 13, 2006 Democracy Out of Anarchy: How Do Features of A Civil War Influence the Likelihood of Post-Civil War Democracy? Mehmet Gurses Department of Political Science University of North Texas Box Denton, TX Phone: Fax: mg0069@unt.edu T. David Mason Department of Political Science University of North Texas Box Denton, TX Phone: Fax: masontd@unt.edu An earlier version of this paper was presented at the Annual Meeting of the International Studies Association, March 22-25, 2006 (San Diego, CA). Word count: key words: civil war, democratization, negotiated settlements.

2 Democracy Out of Anarchy: How Do Features of A Civil War Influence the Likelihood of Post-Civil War Democracy? This study seeks to explore the effects of civil war on post-civil war democratization. We build upon an expected utility model to argue that factors that lead to balanced power relations among the fighting parties will lead to higher levels of democracy. The statistical analysis indicates that civil wars that ended in negotiated settlements are more likely to experience higher levels of democratization, than wars that ended in one side s military victory. Similarly, identity based wars lead to lower levels of democratization. Previous democratic experience seems to decrease post-civil war democratization. We find no support for the argument that high war costs and United Nations peacekeeping forces entail higher levels of democracy.

3 Democracy Out of Anarchy: How Do Features of A Civil War Influence the Likelihood of Post-Civil War Democracy? Empirical research on civil wars has extended beyond the initial concern with the attributes of a nation that put it at risk of civil war onset to the analysis of factors that explain the duration and outcome of civil 1 wars and, most recently, the question of what factors explain the duration of the peace after a civil war has ended. Two strands of research have emerged on this latter question. Research on the impact of international peacekeeping operations has shown that the introduction of peacekeepers does contribute to the durability of the peace. Doyle and Sambanis (2000) highlight the role of peacekeeping and democratization while Fortna (2004) demonstrates that peacekeeping does enable a more durable peace to evolve. A second strand of research has focused on post-civil war institutional arrangements. Hartzell and Hoddie (2003) have shown that in those civil wars that end in negotiated settlements, the more extensive the power-sharing arrangements incorporated in the settlement agreement, the more likely the peace is to endure and the longer it is likely to last. If democratization and power-sharing contribute to the durability of the post-civil war peace, then what conditions make democracy more likely to emerge from the ashes of civil war? We begin with the premise that the prospects for post-civil war democratization are conditioned by the outcome of the civil war itself. Civil wars can end in rebel victory, government victory, or some form of negotiated settlement. Several studies support the proposition that democracy is most likely to emerge from negotiated settlements than from military victories by either the government or the rebels (see Wantchekon 2004; Wantchekon and Neeman 2002). Settlement agreements involve the construction of power-sharing institutions that offer former protagonists the opportunity to compete peacefully for political office and the credible commitments to assure them that they can do so without fear of their former enemy cheating on the agreement and annihilating them with a surprise attack (see Peceny and Stanley 2001; Wood 2001; Walter 2002; Hartzell and Hoddie 2003). By contrast, military victories are less likely to produce post-civil war democratization. Having eliminated their rivals, victorious rebels do not feel much pressure to implement -1-

4 power-sharing institutions. Likewise, victorious governments are not often inclined to do so either, for fear that such institutions would empower the remnants of the rebels support base. This paper focuses on the question of what circumstances are conducive to the emergence of democracy out of civil war. We begin by exploring the impact of civil war outcome (i.e., whether the conflict ended in a government victory, a rebel victory or a negotiated settlement) on the post-civil war regime. Intuitively, one might conclude that the immediate aftermath of a civil war is not a very hospitable environment for democratization to occur. As Roy Licklider (1995: 681) put it, How do groups of people who have been killing each other with considerable enthusiasm and success come together to form a common government? Yet, Wantchekon (2004: 17) notes that 40% of all civil wars that occurred between 1945 and 1993 resulted in an improvement in the level of democracy. Thus, our task is to identify the features of the post-civil war environment that support or retard democratization. To this end, we utilize an expected utility model to develop a theory of post civil war-polity change, from which we derive a series of testable hypotheses. After specifying the research design, data and operationalization of the variables required to test the hypotheses, we present the empirical results from alternative specifications of OLS regression models with pre-war to post-war change in a nation s democracyautocracy scores (from Polity IV) as the dependent variable. We conclude with a discussion of the implications of these findings for the prospects for democratization in the aftermath of civil war. Civil War and Democratization Within the large body of literature on civil war the civil war-democracy link has not yet received as much scholarly attention as questions of civil war onset and duration. Recently, however, some scholars (e.g., Wood 2001; Wantchekon and Neeman 2002) have explored the relationship between civil war and democratization. These studies suggest that inconclusive wars (i.e., wars that do not result in a decisive victory) that are also very costly can lead warring parties to seek a negotiated settlement from which a more inclusive and democratic system can emerge. Earlier works have shown that the longer a civil war lasts, the less likely it is to end in a decisive victory by either side and the more likely it is to end in a negotiated settlement (see -2-

5 Mason and Fett 1996; Mason, Weingarten and Fett 1999). These finding lend support to the proposition that negotiated settlements are a way out of war for protagonists locked in a costly and stalemated civil war. The common conclusion in these studies is that civil war costs and civil war outcome have substantial impacts on the post-civil war political environment. High war costs and long duration reduce both sides estimate of their probability of victory and increase their estimate of the total costs they will have to absorb in order to achieve victory, thereby reducing their expected utility from continuing to fight (in anticipation of eventual victory) versus seeking a settlement. War outcomes, on the other hand, largely determine how post-civil war power will be distributed. Incorporating power sharing arrangements in the post-civil polity should contribute to the development of a more democratic post-civil war polity, which, presumably, should contribute to a more durable peace. Negotiated settlements provide the basis for power sharing arrangements, which, in turn, might evolve into more democratic polities. However, not all civil wars end in negotiated settlements. How does a decisive victory by either the government or the rebels affect post-civil war polity change? How do the direction and the magnitude of post-civil war polity change vary across different civil war outcomes? The balance of power between politically mobilized groups is a primary determining factor of the polity type in a post-civil war state. To model this, we employ an expected utility model of protagonists choice between continuing to fight or stopping. This model has been used to explain civil war outcomes (Mason and Fett 1996; Mason, Weingarten and Fett 1999), civil war duration (Balch-Lindsay and Enterline 2000; Reagan 2002; DeRouen and Sobek 2004), and the duration of the peace following a civil war. The same choice parameters can be used to explain how the prospects for democratization might be enhanced or depressed by civil war. At any given point in the conflict each side has to choose between continuing to fight or stopping. If both sides choose to keep fighting, the conflict continues. If one side stops while the other continues fighting, the conflict ends in victory for the latter party. If both sides agree to stop, the war ends with a truce that can be institutionalized by the negotiation of a settlement agreement. Thus, over the course of a civil war both -3-

6 protagonists gain more information on each other s capabilities that allows them to estimate more realistically their own prospects for victory versus defeat as well as the amount of time that will be required to achieve victory. This same information provides them with the grounds to estimate the prospects for known and even power among mobilized groups in the aftermath of the war. As Przeworski (1991:80-81) has argued, democracy can emerge from bargaining and the outcome of that bargaining is predominantly shaped by whether the relation of the forces is known to the participants when institutional framework is being adopted and, if yes, whether this relation is uneven or balanced, Experiencing a civil war can provide a certain amount of information regarding the capabilities of warring factions, thereby lays the ground for reaching an equilibrium from which democracy emerges. Therefore, we argue that factors that increases the cost of conflict and reduce the probability of a military victory are more likely to generate a post war situation where power becomes relatively even and known. Such post war environment is most conducive to the emergence of democratic institutions by which power is shared by the previously warring factions. Civil War Outcomes and Polity Change Warring factions prefer a military victory to the other possible outcomes (defeat or settlement) because victory yields the greatest utility. However, the costs that a party will have to absorb in order to achieve victory reduce this utility and, at some point, the net payoff from a settlement can exceed net utility from victory. The impact of military victory on the post-conflict balance of power between groups will vary with the amount of costs absorbed by the nation as a whole during the civil war and by each of the contending factions individually. A civil war redistributes both economic and political power within a country. How these resources are redistributed is conditioned by whether the conflict ended in a government victory, rebel victory, or negotiated settlement. A negotiated settlement is the most obvious example of accepting the inclusion of opposition groups into the decision-making process. As has been argued (e.g., Mason et al. 1999; Hartzell 1999; Hartzell and Hoddie 2003) a negotiated settlement is more likely to come after long and costly wars and more likely to be associated with political, economic, and territorial power sharing. Nasr (1990) argues that the first phase of -4-

7 the 1975 Lebanese civil war ( period) coexisted with economic prosperity despite the war. He argues that (1990:5) when the war started, enough public and private economic and financial reserves existed in the country to compensate for the losses and disruption of the first few years. Nasr s argument suggests that, although the first phase of the Lebanese war ended in a negotiated settlement, the nation did not reach the economic limits of its capacity to sustain conflict. The second phase, , however, exhausted the Lebanese economy and consequently the warring factions fighting capabilities. GNP per capita declined from $2,011 in 1982 to $365 in 1988, before the Ta if agreement was signed in The second phase ended with a negotiated settlement that redistributed political power among the several factions. Conditions that make a settlement more likely as the outcome of the civil war should also make post-civil war democratization more likely and more durable. Settlement agreements involve various dimensions of power sharing, including political, military, economic and territorial (Hartzell 1999; Hartzell and Hoddie 2003; Hartzell, Hoddie, and Rothchild 2001). For the settlement to hold and peace to endure, enforceable power-sharing arrangements must be institutionalized in the post-civil war state, and democracy is one institutional form by which this goal can be achieved. Democracy becomes attractive as an alternative to war, not so much because the protagonists have a deep normative commitment to democratic values but because democracy is a way to restore order while controlling the ability of one s rival to revert to political violence. It becomes especially attractive if both protagonists estimate their chances of winning power through democratic elections to be reasonably good (Wanchtekon 2004). A negotiated settlement is most likely to even power balances among the warring parties and lead to more democratic polities in post-civil war states. On the other hand, military victories are likelier to lead to unbalanced and known power relations which will lead to less inclusive institutions. This discussion is summarized in the following hypothesis: H1: the level of democracy should be higher following civil wars that end in negotiated settlements compared to those that end in military victories by the government or the rebels. Not all civil wars end in negotiated settlement. Therefore, it is incumbent upon us to consider how military victory by either the government or the rebels affects the prospects for post-civil war -5-

8 democratization. We first look at what conditions make a victorious government more likely to choose democratic reforms and what conditions reinforce the tendency to build autocratic institutions. The choice of incorporating rebels by building or extending democratic institutions versus resorting coercive measures following a government victory, we believe, is conditioned by the type of regime already in place. Defeating an armed rebellion, for instance, may not necessarily produce a durable peace unless the government either addresses the underlying causes of the conflict through reform or preempts a recurrence of rebellion by establishing the coercive machinery of an autocratic state. Addressing the root causes of a civil war is likely to involve reforms designed to produce improvements in the economic well being of citizens, which is strongly associated with higher levels of democracy. Moreover, political reforms, including democratization, can defuse popular support for rebellion by offering citizens a peaceful means to seek redress of grievances. Civil war, as the most serious threat to an authoritarian regime, however, is more likely to lead to more repressive policies in the post-civil war environment. Civil war is the most serious threat to an authoritarian regime. Once it subdues a rebellion, its reflexive reaction is likely to be a resort to the one institutional capacity that has preserved it against that threat: its coercive capacity. To summarize, the cost of war and the fear of facing the recurrence of war are more likely to push popularly elected governments to open up the system and admit mobilized groups into the polity. However, the same factors might put the survival of an authoritarian regime in jeopardy, leading such regimes to adopt more repressive policies and more exclusive polities. Thus, victory by an authoritarian government victory is least likely to provide conditions conducive to power sharing. H2a: civil wars that end in an authoritarian government victory are less likely to be followed by higher levels of democracy. H2b: civil wars that end in a democratic government victory are more likely to be followed by higher levels of democracy. Rebel victories are more likely to lead to decisive and unquestioned outcome since a rebel victory is defined as either overthrowing existing regime (i.e., Iran 1979, Cuba 1959, and China 1949) or seceding from -6-

9 it and establishing a new sovereign nation state out of a portion of an existing one (Eritrea 1991; Bangladesh 1971). In cases where rebels overthrow the existing regime and establish a new polity, the most common pattern is the destruction of the old ruling-elites and their supporters. Rebel victories, thus, are less likely to be associated with democratization or even ameliorative policies intended to placate previous ruling elites and their supporters. Instead, rebel victories are likely to result in an authoritarian regime that favors its supporters and excludes from the decision-making process the supporters of the old regime. H3: civil wars that end in a rebel victory are less likely to be followed by higher levels democracy. War Cost and Polity Change Civil war is a violent conflict over which faction will control political power through the institutional machinery of the state. The experience of a civil war provides both government elites and rebel elites with new information about the preferences and the capabilities of the warring factions. The onset of a civil war itself demonstrates to government elites that some segment of the population can be mobilized for violent opposition to the state, its policies, and its institutions. As long as state elites calculate that the expected utility from continuing to exclude those groups exceeds the expected utility from admitting them into the polity through power-sharing institutions, incumbent elites will resist pressure to share power and instead continue to fight the civil war. Likewise, as long as rebel elites estimate that their expected utility from continuing to fight in anticipation of eventual victory exceeds their expected utility from conceding defeat or agreeing to power-sharing arrangements with the government, they will continue to fight. However, the experience of protracted civil war is likely to change not only both sides estimates of their chances of achieving victory but also their estimate of their expected utility from agreeing to power sharing as opposed to fighting on to achieve victory and political hegemony. Therefore, the costlier the war, the more accurate both sides estimates of their rival s power will be, and the more likely they are to agree to a democratic post-civil war order: H4: the costlier the civil war, the higher the level of post-civil war democratization Identity Wars and Polity Change -7-

10 Identity wars are likely to have some impact on the post-civil war polity type. As Lake and Rothchild (1996: 71) argue internal wars are particularly difficult to negotiate, largely because ethnic enmities tend to be so deep and the stakes so high. Licklider (1995) also found that identity wars where the warring parties were distinguished from each other on the basis of ethnicity and/or religion were more difficult to resolve through negotiated settlement, and settlements to such wars were less likely to last. Horowitz (1985; 1993) argues that democratic transitions are difficult to negotiate and even more difficult to sustain in ethnically divided democracies. Societies that are deeply divided along ethnic lines tend to produce party systems that exacerbate rather than moderate ethnic conflict. Given this, ethnically based rebels may conclude that they are doomed to permanent minority status in any post-war democracy, with little or no prospect of ever winning enough votes to form a government or even participate in a governing coalition. If they also see themselves as likely targets of a legislative tyranny of the majority, they will be less likely to agree to participate in post-war democratic system and choose instead to pursue their interests through continued violent conflict. Thus, if a civil war based on ethnic and/or religious divisions within a society, the chances of power sharing and inclusive institutions in post-civil war period will be lower. H5: identity civil wars are less likely to be followed by higher levels of democracy. In summary, we expect that civil wars ending in a negotiated settlement are more likely to lead to higher levels of democratization since settlement agreements explicitly redistribute power among the warring parties. Similarly, a victory by a democratically elected government is more likely to pave the way to more inclusive types of institutions. However, victories by authoritarian regimes are less likely to lead to higher levels of democracy since fighting a civil war puts the survival of an authoritarian government in jeopardy. A rebel victory is the most decisive outcome and is likely to lead to power concentration in the post-civil war regime. Hence, rebel victories are less likely to lead to higher levels of democracy. We expect higher war costs provide the incentive for power sharing in the post-war period. Finally, identity wars (ethnic and/or religious) wars are less likely to be followed by higher levels of democracy since such wars are likely to lead to exclusionary institutions in post-civil war states. -8-

11 Research Design The focus of this study is on civil wars that started and ended during the post-world War II era. We use the Doyle and Sambanis (2000) data set, which includes all wars that started after 1947 and terminated in or before Doyle and Sambanis (2000) define a conflict as a civil war if it meets the following criteria: (1) the war has produced more than 1000 battle deaths; (2) the war represented a challenge to the sovereignty of a recognized state and occurred within its boundaries; (3) the state is involved as one of the main combatants; and (4) the rebels were able to mount an organized military opposition to the state and to inflict significant casualties on the state. 2 Doyle and Sambanis data set classifies four outcomes: (1) military victory by government; (2) military victory by rebels; (3) formal negotiated settlement and (4) informal cease-fire/truce. We excluded the cases that ended in truce since no conclusive argument can be made based on the outcome of such cases. Inclusion of such cases may lead to measurement validity problem since the impact of a civil war cannot be measured 3 if warring factions live in separate territories with no conclusive outcome. The unit of analysis is the post- civil war year at five and ten years following the end of a civil war. This includes those cases that experienced another civil war before a 5 and/or 10 year period following the end of a conflict. Dependent Variable The dependent variable is the change in the level of democracy in a post-civil war state. One cannot accurately analyze the effect of a civil war on the post-civil war polity type except in comparison to the prewar pre-war polity. To measure the impact of a civil war on the change in the level of democracy in post-civil 4 war states, one should take into account the level of democracy prior to when the conflict started. Statistically speaking, we treat a civil war as a treatment and measure its effect on regime type following the end of a civil conflict. Hence we subtracted the level of democracy at one year before the conflict started from the level at 5 5 and 10 years following the end of that conflict. Doyle and Sambanis (2000) also measure post-civil war democratization at 5 and 10 years after the conflict ended, but they did not consider those scores relative to the pre-war level of democracy. We used the polity2 variable from the Polity IV data, which is the combined -9-

12 autocracy-democracy scale that ranges from -10 to +10. We rescaled it by adding 10 to each nation s score to give the same 21-point scale but with a range from 0 to +20 (0 = maximum autocracy, 20 = maximum democracy). Rescaling the variable in this manner does not change the results but simplifies the analysis by making all scores positive. The dependent variable, polity change, was constructed as follows: the rescaled polity2 score at 5 (or 10) years after the civil war ended was subtracted from the rescaled polity2 score at one year before the conflict started.. The dependent variable, polity change, ranges from negative 15 (Laos ) to positive 14 6 (Mozambique ). Of the 90 cases that we are able to analyze at year 5 following the end of a civil war, 31 cases ended in negotiated settlements, 22 wars ended in rebel victory, 32 cases ended in authoritarian government victory, and 4 of them ended in democratic government victory. Among these, 26 cases experienced a negative change in their democracy scores, 23 had no change, and 41 cases experienced a positive change at year 5. Almost half of the cases that saw a positive change at year 5 ended in negotiated settlements and only 1 of them ended a democratic government victory. The remaining cases that saw a positive change ended in authoritarian government victory or rebel victory. Of the cases that experienced a positive change at year 10, one-third ended in negotiated settlement and the rest ended in either authoritarian government victory or rebel victory. We checked for the possibility of regimes that are located near the extreme boundaries of the Polity score. That is, a highly autocratic regime (i.e, polity2 approaches -10) cannot have a large negative score on our dependent variable nor can a highly democratic nation have a large positive score on our dependent variable because both types are near the extreme limit of the polity2 scale. This constraint on the dependent variable could generate misleading results. The descriptive statistics reveal that this is not the case. There are indeed a substantial number of highly authoritarian cases that experienced a substantial change following the end of civil war, but those changes were positive in value (i.e., they became more democratic). To clarify to the point, there are 26 cases that were located near the lower boundary of the Polity scale (between 7 and 10) that experienced some sort of positive changes in their levels of democracy at year 5 following the end -10-

13 of the war. The prewar polity score for Mozambique, for instance, was 8. The score increased by 14 points on the Polity scale at year 5 following the end of the war (+6). Similarly, Nicaragua moved from 8 to 1 following the end of the civil war. Paraguay and Chad, on the other hand, moved from 9 to 5 and 4 respectively, at year 5 following the end of civil wars. We also checked the cases that are located at the upper boundary of the Polity scale. There are only 4 cases that had a score of +10 prior to the war and none of these countries experienced any change in their polity scores following the end of the war. Further, we estimated the models with these cases omitted and obtained very similar results. The descriptive statistics suggest that most of the cases under analysis experience a change in their regime types after civil war. What remains to be seen is what accounts for such change, a task we undertake in the analysis section. Independent Variables War Outcome: Four dichotomous variables were defined to denote four war outcomes; negotiated settlement, authoritarian government victory, democratic government victory, and rebel victory. To test for the impact of a negotiated settlement versus a military victory (hypothesis 1), we generated a dichotomous variable equal to 1 if a war ended in a negotiated settlement, 0 otherwise. To test for the impact of authoritarian government victory (hypothesis 2a) we generated a dichotomous variable equal to 1 if the war ended in government victory and the pre-war polity2 score was less than 6 on a scale of 10 to +10, 0 otherwise. To test for a democratic government victory, we generated a dichotomous variable equal to 1 if the war ended in government victory and the pre-war polity2 score was greater than 5 on a scale of 10 to +10, 0 otherwise. War Cost: The war cost is operationalized as war duration because longer civil wars tend to accumulate higher casualties and deplete economic resources (Murdoch and Sandler 2004). Civil war duration is measured in months and logged for computational ease. This variable ranges from 1 (minimum duration) to (maximum duration) with mean equal to months. When we included casualties in the model (not reported), the effect was not significant and the other results were not altered. -11-

14 Identity War: Finally, to test for the impact of identity war we defined a dichotomous variable as equal to 1 if a war was fought on ethnic and/or religious lines, 0 otherwise. The data for the independent variables come from Doyle and Sambanis (2000). Control Variables Economic Development: The relationship between economic development and democracy is wellestablished, dating back to Lipset s (1959) seminal work. We used the log of GDP per capita at years 5 and 10 following the end of a civil war to control for the impact of economic development on democratization. 8 The GDP data (constant 1995 international $) come from the World Development Indicators. Values range from $73 (minimum) to $22, (maximum) with a mean of $1,640 at year 5 and from $ 69 to $ 6,041 with a mean of $1,081 at year 10. Peacekeeping Operation: A dichotomous variable was defined to capture the impact of the UN peacekeeping operations percent of the cases had UN peacekeeping at year 5 and percent had a peacekeeping operation at year 10 (Doyle and Sambanis 2000). Previous Democratic Experience: Previous democratic experience is likely to have some effects on the post war regime type. More specifically, countries with previous democratic experience are more likely (1) to adopt a democratic regime in the first place and (2) to see higher levels of democracy in post-civil war era. A dichotomous variable was defined to control for such experience. This variable was coded 1 for countries that had a democratic regime (polity2 > 5 on the unconverted scale of -10 to +10) in the period before the war started, and 0 otherwise. Ongoing War: Having an ongoing civil war is likely to have substantial impact on the level of democracy in a country. To control for this we defined a dichotomous variable equal 1 if there was an ongoing war at the time the dependent variable was measured, 0 otherwise. Of the cases analyzed in this paper, 8 cases had an ongoing war at year 5 and 8 cases at year 10. Ethnic Fractionalization: To control for the effect of ethnic fractionalization we used Vanhanen s ethnic heterogeneity index, which measures racial, religious, and linguistic divisions within a country. It ranges from -12-

15 4 (minimum heterogeneity) to 144 (maximum heterogeneity). We obtained these data from Doyle and Sambanis (2000). Population: The effect of the population size may be twofold. A large population may increase the conflict within a society by providing rebels with a large poot of potential recruits and by straining the state s capacity to deter and control dissident activity. Moreover, a large population may hinder the development of a democratic system due to the problem of coordination between numerous mobilized groups. The log of population was used to control for the effect of population on the level of democracy. Method The dependent variable is the change in polity scores at 5 and 10 years following the end of a civil war. 9 Due to the nature of the dependent variable the hypotheses were tested by estimating OLS regression with robust standard errors clustered by same country observations to relax the assumption of independence of the observations. Analysis To test for the impact of negotiated settlements versus military victories, we estimated the model without the military victory (by either side) variables. These results are shown in the Model I of Table 1. Model II unpacks military victories and compares the impact of victory by authoritarian governments, democratic governments, and rebels, respectively, to that of negotiated settlements. The same analysis was conducted for the year 10 following the end of a war. The regression results are reported in Table 1. [Table 1 Here] The results suggest that negotiated settlements, as hypothesized, lead to increases in the higher levels of democracy following the end of a conflict: on average, if a war ends in a negotiated settlement that country experiences a 3-point increase in its polity score relative to its pre-war polity score. The p-value for this variable is.04 at year 5 and.13 at year 10. However, given the small sample size (N=90 at year 5; and 65 at year 10), we believe that the findings confirm our expectation regarding the relationship between settlement and polity change. Further, because of the missing data on some of the independent variables, the number of -13-

16 cases that ended in a negotiated settlement reduced substantially at year 10 (from 32% to 21%). The results support the argument that decisive victory by either side will lead to lower levels of post-civil war democracy. Authoritarian government victories and rebel victories decrease the level of democracy by more than 3 points at year 5 and year 10 relative to the pre-war polity. The results for identity wars are in the expected direction and substantially and statistically significant. Identity wars decrease the level of post-civil war democracy by almost 3 points at year 5 and by more than 4 points at year 10. This finding is compatible with the current literature on ethnically divided societies. In deeply divided societies political parties are more likely to be organized along their ethnic lines. As Horowitz puts (1985:196) election is a census, and the census is an election. Such divisions are more likely to lead to exclusive polities following the end of a civil war between ethnic/religious groups. The animosity between powerful groups is not likely to go away if a war is fought based on identity lines. An identity war, hence, will not lead to inclusive polities, in which previously excluded groups can be represented. We found no support to the war cost hypothesis. Previous research (Mason and Fett 1996) has shown that war duration has a substantial impact on war outcome: the longer a war lasts, the more likely it is to end in a negotiated settlement. Since negotiated settlement is associated with higher levels of post-civil war democracy, it may be that whatever variance in post-civil war democracy is explained by the duration of the conflict is already explained by the outcome variable, negotiated settlement. In other words, the effects of war costs on post-civil war democratization may be indirect: war costs make negotiated settlements more likely, and it is the settlement outcome that increases the level of post-civil war democratization. We estimated the models with other measures of war costs, including casualties, and the results were similar to those reported here: war costs (however measured) do not seem to affect the level of post-civil war democratization directly. It was hypothesized that democratic government victories will lead to higher levels of democracy. The finding on democratic government victory does not lend support to our hypothesis. Given the direction of the hypothesis, we fail to find support for a positive and significant relationship between a government victory and higher levels of post war democracy. Nevertheless, we also belive that there is a need to go beyond -14-

17 statistical results and investigate what accounts for such negative and significant finding. Notice that this finding is reinforced by the negative and significant the previous democratic experience variable (the p- 10 value for the previous democratic experience=.03 and 0.13 (two tailed) at year 5 and 10 respectively). That is, countries with a previous democratic experience tend to become more authoritarian following a civil conflict. Therefore, despite the lack of support for our democratic government hypothesis, there is more to be said about such counterintuitive finding. One possible explanation is that the majority of countries with previous democratic experience are new democracies. A brief democratic tradition impaired by a civil war may induce the ruling elites within a country to conclude that an inclusive decision-making process is not suitable for their societies. The Nigerian case illustrates the relationship between previous democratic experience and post-civil war polity change. Nigeria gained its independence in 1960 and remained democratic until one year before the civil war started. The country experienced a sharp decline in its polity score in the post-civil war years. In 1965 the polity2 score was 7 whereas in 1975 it dropped to 7. The country, however, re-established its democratic system in 1979 and the polity2 score rose to 7 again. This was followed by another civil conflict, which ended in Nigeria s polity2 score dropped to 5 at year 5 (1989) and to 7 at year 10 (1994) following the end of the second war. Uganda experienced a similar pattern. After four years of democratic experience the country had a civil war in The polity2 score dropped from 7 to 7 at year 5 following the end of the internal war. There are some common elements in these cases, which may be useful in explaining this puzzle. Both countries are new states (both gained their independence in the 1960s from the UK) and adopted democratic regimes after they became independent. It is likely that these countries adopted democracy mainly due to the identity of the colonizer, the United Kingdom, not due to a genuine domestic process that produced inclusive governance. The colonial powers, however, sow the seeds of internal conflict within the colonized societies through creating artificial boundaries and dividing the colonized peoples, on the one hand, and imposing their political system, parliamentary democracy, on the other. The contradictions within these -15-

18 societies ended up in bloody internal conflicts. As a result the winning side in the civil war abandoned the democratic polity, justifying post-civil war authoritarianism as necessary to prevent a relapse into civil war. Finally, the results do not lend support to the argument that UN peacekeeping missions increase the level of democracy. This variable has some positive effect on the post-civil war polity yet it is not significantly different from zero. This conflicts with the findings of Doyle and Sambanis (2000). They found that the UN involvement helps sustain peace and support democratization. This effect, as they argue, is more visible and stronger for the missions with extensive civilian functions, including economic reconstruction, institutional 11 reform, and election oversight [multidimensional mission] (2000: 791). Our findings diverge from theirs in part because Doyle and Sambanis (2000) use a dichotomous measure of democracy and more importantly do not control for the pre-war level of democracy. We believe that the pre-war regime type should have substantial influences on the post-war regime type. Hence, it needs to be incorporated into the model to come up with more valid and confident results. The literature on civil indicates a possible collinearity between some of the variables (e.g., war outcome, war duration, UN peacekeepers, war type) used to estimate our models. We checked for multicollinearity by th using the variance inflating factor (vif). The vif measure is based on regressing a j regressor on the others 2 to determine the amount of variance explained by the variable of interest. Then R j is used to calculate the 2 2 vif, which is obtained as follows: 1/1- R j. In this instance, the larger the R is, the more severe the multicollinearity is. As a rule of thumb a value of 10 and higher for the vif indicates severe collinearity (Gujarati 1995:339). The vif values for our models have a mean value of 1.61 and do not exceed 3 for any regressor. Furthermore, we estimated the models by excluding some of the variables to see the estimated effect of the variable of interest changed. For instance, dropping the war outcome variable(s) yields significantly positive (one-tailed) estimates for the UN peacekeeping variable (results not shown). However, one should keep in mind that dropping a variable from the model might cause specification bias. In addition to low values for the vif, multicollinearity does not violate any regression assumption and still produces unbiased and consistent estimates. Therefore, we conclude that our estimates are not a by-product of -16-

19 multicollinearity and the effect of war outcome seem to wash away the effect of UN peacekeepers and war cost variables. Conclusion We have explored the effects of experiencing a civil war on the post-civil war polity. The point of departure was how an internal conflict redistributes economic, social, and political power in a society through war costs and war outcomes. Solving the puzzle between civil wars and post-civil war polity is important since a substantial number of developing countries have had civil wars or are experiencing them. The results provide support for the argument that power distribution is crucial for a democratic regime to be established. Civil wars can shift power balances within a society, making one group s dominance less likely. Hence, having an internal conflict may not necessarily have negative effects on a country s polity. Indeed, as Wantchekon (2004) argues, democracy may arise from anarchy, but that outcome is contingent upon how that anarchic situation ends. Wars ended through negotiated settlements are more likely to provide the basis of power sharing and, hence, lead to more inclusive polities in post-civil war environment. This study contributes to the current studies (e.g., Hartzell 1999; Hartzell and Hoddie 2003; Wantchekon 2004) by testing for the relationship between civil war features and post-civil war polity type. The main finding is that civil wars may lead to more inclusive polities by redistributing power in a country. However, we did not find support for the argument made by some scholars (e.g., Doyle and Sambanis 2000; Walter 2004) that peacekeeping operations lead to higher levels of democracy. Although these studies primary concern is not post-civil war democratization they argue that the presence of peacekeeping forces provide the basis for more democratic regimes in post-civil war countries. The findings of this analysis suggest that internal dynamics are more important than outside intervention as far as democratization is concerned. We believe, how the war ended should have an impact on the post-civil war political environment and that is what both the theoretical works cited earlier as well as a body of empirical works suggest. More specifically, since almost all peacekeeping operations are introduced following negotiated settlements (not military victories), it could be that the peacekeeping effect Doyle and Sambanis found is actually a war outcome effect. -17-

20 Of course, this study is by no means is conclusive. The findings should be tested against alternative data sets and measurements. Further, the effect of a democratic government victory and previous democratic experience needed to be analyzed more thoroughly. Common sense suggests that a victory by a democratic government and a previous democratic tradition should lead to higher levels of democracy. Nevertheless, both the statistical analysis and an examination of cases indicate a negative change following a democratic government victory and a previous democratic experience. We have employed several alternative measures of the latter variable to check the robustness of the finding. Of the four democratic government victories only one case experienced a positive change following the end of the war (India). Others either had no change or became more authoritarian after the war ended. One possible explanation is that, because a substantial portion (2/3) of civil wars in countries that had a previous a democratic experience ended in one side s military victory, a civil war might alter the preferences of the victor concerning how to rule. That is, civil wars that end in one side s victory may dramatically alter the balance of power between the fighting groups, on the one hand, and increase the perceived threat posed to the regime, on the other. Given that most of the countries that experienced civil wars are new states with relatively weak democratic experience, one side s military victory may lead the ruling elites and/or the victor to conclude that the previous democratic regime was not appropriate for their country. Victory gives them the power to repress residual or potential challengers in the post-civil war environment, resulting in a less democratic polity. The findings indicate that democratization in post-civil war countries may have a different dynamic. Even after controlling for socio-economic factors, such as ethnic heterogeneity, economic development, and population, power sharing seems to lie at the core of higher levels of democracy in post-civil war countries. Future research is necessary to check for the robustness of the finding by employing finer measurements and alternative data sets. In conclusion, this study suggests that experiencing a civil war may lead to more inclusive polities if it serves to even the balance of power between contending groups in a country. Power balance is more likely to bring about more democratic polities. The insignificant war cost (measured as war duration) indicates that -18-

21 unless warring factions settle on the terms of peace, the chances for a more democratic polity are low. Thus, the international community can change the post-civil war political environment by not only sending peacekeeping forces to the field but also by providing the ground for negotiated settlements through negotiations, arbitrations, and mediation. -19-

22 1 On the onset of civil war, see Fearon and Laitin (2003). On the duration of civil wars, see Balch-Lindsey and Enterline (2000), Fearon (2004), DeRouen and Sobek (2004). On the outcomes of civil wars, see Mason and Fett (1996), Mason, Weingarten and Fett (1999). 2 Doyle and Sambanis coding does not require 1,000 battle deaths annually. They use 1,000 deaths threshold for the entire period as long as the war produced at least 1,000 deaths in any single year. This more relaxed (than the Correlates of War) standard means they include some conflicts that would not be included in the COW data set. Hence, it provides a better tool to measure the impact of a civil war on post-civil war polity type. See Sarkees (2000) on the Correlates of War inclusion criteria and Sambanis (2004) on the relative merits of different inclusion criteria. 3 For instance the levels of democracy in South and North Korea, and the Turkish and Greek Cyprus, are not likely to be a factor of civil wars that occurred between these countries. 15 cases (12.10%) of the sample ended in truce. 4 For cases where rebels successfully seceded (e.g., Bangladesh 1971, Eritrea 1991) the pre war polity of the country whom rebels seceded from was used. 5 Where the Polity Score at one year before a conflict was not available, we used the most recent polity score Endnotes These values are for year 5 following the end of the conflicts. Using casualties as a measure of costs did not lead to any substantial change in the results. Accessed at on March 25, We used gdp per capita (constant 2000 international$) data for Ethiopia, Israel, Kenya, and Zimbabwe. Accessed on March 31, For cases where these data were not available we used the gdpen variable -20-

23 in the Fearon and Laitin (2003) data set. The latter data (1985 constant USA $) originally come from Penn World Tables. 9 This study argues that the impact of a civil war on a country s polity type can be observed better in the years following the end of the conflict. Since any institutional change, especially such as the one analyzed in here, is likely to follow a political/economic shock, it is more likely that a change followed by the end of that shock (civil war in this study) will take place right after it. Over the time this impact may or may not go away. However, changes over time may not necessarily be a consequence of a shock that a country experienced say 25 years ago. Hence, we believe the first 5 years after the end of a civil war reflects the change theorized here better. Hegre et al. (2001) also use change in the Polity score as a predictor of civil war onset. The change at year 10 was used as a check. Polity change (dependent variable) at years 5 and 10 was the same for the majority of countries analyzed in this study. A substantial portion of those cases that have different polity change at years 5 and 10 were countries that had subsequent wars. The model controls for the ongoing war by including a dichotomous variable equals 1 if there was an ongoing war at the time the dependent variable was measured, 0 otherwise. 10 The model was tested with different measurements. To control for the effect of previous democratic experience we used (1) the mean polity score for 5 years before a civil war started; (2) total number of years in which a country had a polity score higher than 6; (3) the polity score at one year before conflict started; and (4) moving average for the 5 years before a conflict initiated to check the robustness of the results. All gave similar results. 11 Using the untype variable, which ranges from 0 (no mission) to 5 (multidimensional peacekeeping), in the Doyle and Sambanis data set yielded similar results. -21-