Out of the frying pan into the fire?

(1)

Out of the frying pan into the fire?

An Event History Analysis of Democratic Breakdown after a Civil

War

The global trends of political regimes have always been an important field of study in political science. However, much of the focus has solely been on democratization, instead of democratic breakdown. With the current international political climate in which certain democracies seem to reverse, we analyze the conditions under which democracies break down. By doing so, this thesis contributes to a more comprehensive knowledge of the decay of democracies by performing a cox proportional hazards test into democratic breakdown after a civil war, while controlling for certain institutional, economic and civil war covariates. We build on the extensive civil war and

democratization literature that is available and carefully apply these findings to democratic breakdown. Through a dataset that we build, we apply a number of covariates to democratic

breakdown. The most important covariates are a number of dummy variables for each of the different civil war outcomes, namely a rebel victory, a government victory and a peace treaty. Our analysis shows that peace treaties have a better effect on democratic survival rates than other forms of civil war settlement. However, none of the civil war settlement effects remain statistically significant when we control for our other covariates. Our main finding lies in the significance of a recurrent democratic breakdown. This phenomenon has hardly been studied and requires more academic attention.

Master thesis Tom Dymanus

S4329090 13209 words

(2)

Chapter 1: Introduction

1.1 background

The political system of states has been an internal matter ever since the establishment of the treaty of Westphalia in 1648. Yet, scholars have always been interested in global patterns that explain the spread of different political systems across the globe. More recently, the trend has been to view the zeitgeist in terms of ‘waves of democratization’. Samuel Huntington

famously introduced this term in order to argue that after a first and second wave of democratization followed both world wars, a third one was occurring at the end of the twentieth century (Huntington, 1993). He argued that this wave was caused by communist regimes that were crumbling and were being replaced by democratic regimes (Huntington, 1993). Thinking about democratization in terms of ‘waves’ makes scholars hopeful, because it implies that these waves add new democracies every time they occur. More recently the debate has focused on the extent to which the Arab Spring can be considered as a wave of democratization or whether we remain too optimistic about global developments, as was probably the case with Huntington. Underlying to these debates are more fundamental

questions about the direction in global regime trends. Answering these questions in the wrong way can lead to false optimism regarding the global progress of democracy. In order to correctly assess the developments that we are witnessing, we need to distinguish between different mechanisms that can sustain democracy like economic growth, the availability of natural resources and the domestic balance of power (Joshi, 2010). Even though a lot of research has been devoted to this subject, the exact mechanisms behind the workings of democracy remain unclear. However, in order to even begin to answer questions about the global direction of political systems, it is important to look at both sides of the coin. While more classic coup d’états and election day fraud no longer seem to be a functioning method of regime change, we have still seen a number of democracies backslide, recently through different forms of democratic decline. Seizing power through slow coups that promise to restore democracy in the long run, strategic manipulation of the elections without obvious violations of the law and weakening checks on executive power have all increased

substantially since the 1990’s (Bermeo, 2016). This goes to show that just because there are fewer dictators bluntly seizing power does not mean that there are no democracies breaking down. The ways to seize power are just less obvious and stretch over a longer period of time. Therefore, both democratization and the breakdown of democratic regimes are important to research. Doing so will prevent us from misrepresenting the global direction of political regimes.

1.2 Research question

Where some of the causal mechanisms for democratization can be described as a dubious proposition at best, they are downright non-existent when it comes to democratic breakdown. Many factors that could negatively impact a country’s democracy have not even been

researched at all. One of these potential factors is whether a country has had a civil war in its past. Civil wars are often highly disruptive to a country’s political system and might pose continuing problems during peacetime if not settled properly. This means that the disruptive nature of a civil war does not necessarily end with the violence. Numerous scholars have

(5)

already argued that different civil war endings can have different and even disruptive effects on democracies (Gurses & Meeson, 2006; Wantchekon & Neeman, 2002). If the disruptive nature of a past civil war is not addressed, it might play a role in a country’s democratic breakdown, even if this democratic decline happens years later. Our main research question is therefore:

To what extent does a civil war in a democracy’s past affect the likelihood of a democratic breakdown?

The risk of democratic breakdown after a civil war has not received any academic attention so far. Even when it comes to the relationship between democratization and civil war, different studies all point towards different and often contradictory mechanisms that they deem plausible. This is mostly caused by different effects on the long and short term, different measurements between studies and different methods that are being used. Unraveling which studies are correct is an important goal of this study, since it will clarify how the different mechanisms within democracy work. This is important for the research into democracy, since it can help the advancement of universal principles for the stability of democratic regimes. Different scholars have found causal relationships between a certain civil war outcome and the chances of democratization (Joshi, 2010; Gurses & Mason, 2000; Wantchekon & Neeman, 2000). We will perform a similar research with an event history analyses into how a country’s history of having a civil war influences democratic breakdown, even after the war is over and a country has democratized. The study will make use of the Polity IV dataset that

encompasses all the civil war cases between 1945 and 2015. Different civil war characteristics might have a different effect on short, mid and long term democratic breakdown, because underlying causes of the initial conflict are not properly addressed in the termination of the war. A civil war has different characteristics that we will research. Most importantly, we will see to what extent the civil war endings have a different effect on democratic survival. Another example is how the duration of the civil war could influence the extent to which it later affects democratic survival. We draw inspiration from the democratization literature regarding civil wars in order to gain a clearer understanding of the workings of democracy and civil war.

1.3 Scientific relevance

Establishing such a relationship between a country’s history and the chances for democratic breakdown helps us in understanding why democracies fail. This relationship deserves more attention, since it might cause countries with a similar civil war outcome to follow the same trends in democratic breakdown. Our research could also help by encouraging new research into democratic breakdown in relation to other factors that might prove important. This study also contributes to the amount of academic knowledge that we have on civil wars and their consequences. Most of these studies have been performed while using a very limited

timeframe. By expanding the timeframe, we can get an understanding of the long term effects that civil wars can have on a country’s stability.

1.4 Societal relevance

It is important to further study this relationship, since the United Nations (UN) and other international actors devote a lot of economic and institutional aid to making sure that both a

(6)

civil war is ended as quickly as possible and that countries with a past civil war do not slide back into violence. If certain civil war outcomes are more likely to result in continued unrest, they would become less feasible for countries and the UN to use. This argument also extents to democracies, with a lot of foreign aid being devoted to upholding democratic values in fragile states. There is also the recurring idea that countries need to be ‘ready’ for democracy and the question of whether it is too much based on ‘western principles’ to be successful in other parts of the world. This speaks to the idea that countries need to have a certain level of development or education before they can successfully sustain a democracy. By using a number of control variables, among which institutional and economic variables, this research will contribute to answering these type of questions.

1.5 Thesis outline

We continue this thesis with a theoretical chapter in which we expand on the existing literature. Since we are trying to link democratic breakdown to civil war factors, we will discuss both. It is important for this research to get an idea of the different features of civil war settlement mechanism and how they can later impact stability. The third chapter discusses the method and clarifies the different variables that we will be using in this thesis. Afterwards, we will perform the cox proportional hazards test and present its results. In the fifth and final chapter, we will discuss the findings and limitations of this thesis while ending with some recommendations for future studies.

(7)

Chapter 2: Theory

2.1 Democratic breakdown literature

The way democracies appear and disappear is an important theme in understanding the development of political systems across the globe. To emphasize democratization in this regard seems logical, because the general direction is hoped to be towards the establishment of democracies and the expansion of civil liberties. However, another crucial aspect of democracy lies in understanding how established democracies are able to maintain their status as a functioning democracies. Some democracies have been stable throughout long periods of time, while others have descended towards a more authoritarian system. So far, only a few scholars have run tests studying democratic breakdown by taking into account different variables that could account for a breakdown. Characteristic for these studies, and for political science in general, is that there are no laws which have been proven yet. Nevertheless,

different scholars have found possible mechanisms that could account for an increased likelihood of democratic breakdown. Maeda focusses on the effect that having a presidential or parliamentary system has on democratic breakdown. He finds that presidents are more likely to become authoritarian than prime ministers (Maeda, 2010). He argues that this could be caused by the large emphasis on the separation of power within a presidential system. The tidiness of systems with strong separation of power may tempt presidents in seeking

unconstitutional solutions to achieve their goals (Ibid.) This effect could be further

strengthened by the fact that losing presidential candidates are not elected to office, which could be another incentive to take matters into your own hands (Przeworski, Alvarez, Cheibub and Limongi, 1996). Maeda has also found that the risk of democratic breakdown is

negatively related to both economic development and the economic growth rate (Ibid.). However, Mainwaring and Perez-Liñán focus on Latin America and contradict the widespread claim that the level of economic development is crucial for the survival of democracy in this region (2013). This goes to show that there is still a lot of debate about the consequences of specific economic factors on democratic breakdown. However, the idea that the functioning of an economy influences the sustainability of democracy is one of the few claims that is generally accepted (Maeda, 2010; Blattman & Miguel, 2010 & Joshi, 2010). A flourishing economy seems to aid democratic survival, but is by no means a guarantee. Instead, Mainwaring and Perez-Liñán have coded a large number of political actors on a scale from radicalism to a normative preference for democracy and found that countries with less radical policy preferences are far less likely to break down (Ibid.). Central in their argument is the idea that popular mobilization threatens dominant elites up to the point that they become opposed to democracy in general (Ibid.). This implies that elite approval is central to

maintaining certain levels of democracy.

However, there are still a number of different plausible explanations for democratic

breakdown that have not been studied at all. As we have demonstrated above, the emphasis of previous research has been on the effect of economic and institutional variables. So far, these variables are by no means sufficient in explaining the vast number of cases of democratic breakdown. Certain studies do indicate causal relationships, but are often contradicted by other research. The emphasis on economic and institutional variables has resulted in very few

(8)

studies that actually look into large scale cleavages which affect a country’s demise of democratic levels. Mainwaring and Pérez-Liñan (2013), for instance, acknowledge the influence of ‘policy radicalism’, but fail to address the underlying causes of said radicalism. Exploring the origins of these deep-rooted divisions within a country might be part of the solution for democratic breakdown. One of the possible origins of such divisions within a country can be traced back to a civil conflict in a country’s (recent) history.

In this study, our main focus will be on researching the extent to which a country’s civil war in the past could still be a factor in the collapse of its democracy later on. However, we will still need to use democratization literature as a starting point, because the only existing studies that link civil war to democracy are about the process of democratization. Democratization literature could give us more insight in the relationship between democracy and civil war in general. We will also make further use of the existing democratic breakdown literature. It seems likely that the two are related, since the absence of variables that aid in democratization could also play a role in the breakdown of democracies. However, there is very little

consensus in democratization literature, especially when it comes to the causal mechanisms between democratization and civil war. It is thus important to study the different arguments that are made for the causal relationship between civil war and democracy more closely in order to apply them to democratic breakdown.

2.2 Civil wars

Recent times have brought a significant increase in the number of civil wars up to the point that all of the 15 major armed conflicts listed by the Stockholm International Peace Research Institute for 2001 were internal (SIPRI, 2002 in: Collier & Hoeffler, 2004). In order to fathom the full impact of civil wars, it is also necessary to get an idea of just how common it is for a country to experience a civil war at some point in its existence. For instance, the 1990’s showed a peak in civil war occurrence with over 20 percent of the countries globally being embroiled in some type of civil conflict at that point in time (Blattman & Miguel, 2010). Most of the civil war definitions necessitate a conflict between government and some type of rebel organization that is internal to its sovereign territory and usually has a threshold of more than 1000 deaths over the course of the war (Sambanis, 2004). It remains difficult to accurately point out the causes of civil war, because there are so many. In general, civil wars occur when grievances are sufficiently acute that people are willing to engage in violent protest (Collier & Hoeffler, 2004). These grievances can have different origins like poverty, ethnic divisions or other cleavages. A combination of different factors is often at play, which makes it hard to pinpoint an exact cause.

As we have mentioned during the introduction, a civil war can prove highly disruptive for a country’s political processes. Not only does it often cause legislative bodies to be shut down completely, it can also create huge cleavages within society. There are also scholars who point out that civil wars can aid the technological and institutional development, as was the case in East Africa and Southeast Asia (Blattman & Miguel, 2010). Since the consequences of civil war are not clear-cut, the exact consequences of a civil war and a country’s best way to recover from it remain the subject of extensive debate. The key goal for both scholars and state officials so far has been to achieve stability in order to prevent the renewal of violence. There is a huge body of civil war literature that studies every major aspect of it, like what the

(9)

best approach is to establish democracy after a civil war and how civil wars have changed throughout time. This is also where we will draw our inspiration from for this thesis. In her studies, Toft shows that the number of decisive military victories from civil wars has drastically decreased from around eighty percent in the 1940’s throughout the 1980’s to only 41 percent in the 1990’s and later (Toft, 2010). This changes the dynamics of a post-civil war situation, because there are often more parties that have to be taken into consideration, since there is no sole victor. Scholars have also researched the effect of different variables on state stability after a civil war, such as the characteristics of the war, the characteristics of the peace agreement and the civil war outcome. In general, there are two categories of variables here. The first category has to do with the environment in which a civil war is resolved. This ‘environment’ category mostly refers to the type of regime that is present in a country and the regional and international circumstances of it, like the stability in neighboring countries and whether the civil war ended during the cold war. The second category on the other hand, has to do with the impact of the settlement (Hartzell, Hoddie & Rothchild, 2001). This category is about the different civil war outcomes and their chances for democracy. Both of these

categories can have their own impact on a country’s later stability and chances for a sustainable democracy.

2.2.1 The environment of a civil war

When it comes to the former category regarding the environment in which a civil war has ended, Hartzell, Hoddie & Rothchild have tested the degree to which the previous regime type before a civil war matters. They found that if the last regime type before the civil war was democratic, the chances of a failed negotiated settlement decrease by 90 percent (2001). This implies that former democracies tend to result in more stable peace settlements. When it comes to democratic breakdown, however, former experience with democracy also implies that a country has already descended from democracy into civil war. This might make another instance of democratic breakdown more likely.

Another important part of the environment of a civil war has to do with the intensity of the civil war. For instance, there is discussion about the effects of the cost of war on

democratization. Huang argues that the increased loss of human life will further impede the democratization process, because it affects the trust and compromise needed for democracy to be established (2008). Gurses and Mason oppose this view by arguing that a longer duration of war will create a form of war fatigue which makes a democratization process more likely (2008). This idea of ‘war fatigue’, implies that people just get sick of war after a long period of it and that this somehow benefits the likelihood and the duration of peace. This could be the case because of the positive effect of a negotiated settlement on democratization, as Mason and Fett argue that the longer a war lasts, the greater the chance of a negotiated settlement (1996). This links with the finding by Hartzell, Hoddie and Rotchchild that conflicts of low intensity which have endured for a longer period of time are more likely to result in a durable negotiated settlement (2001).

(10)

2.2.2 Consequences of a civil war settlement

The second category we just discussed, the effects of different civil war settlements, is based on the idea that the different ways a civil war can end have different mechanisms towards or away from democracy. The most important ways a civil war can end are either military victory by either side, or a negotiated settlement. Although there is a lot of discussion about the different consequences of them, it is important to note that there is no real difference in terms of casualties between wars that end in negotiated settlements and military victories (Licklider, 1995). This implies that the circumstances of a war do not determine the way a civil war is settled. Whether a different civil war settlement means different chances for democracy is still the subject of lots of debate. Gurses and Mason argue that post-civil war democratization is far more likely to occur if the civil war outcome has resulted in a balance of power between warring parties (2008). This seems to speak for a negotiated settlement, because the parties have at least recognized the other in such an agreement. Wantchekon and Neeman come to the similar conclusion that diversity is the key factor in democratization, because the range of political preferences makes that each group assesses their own chances of winning as high enough (2002). This emphasis on pluralism in democracy’s success seems to be a recurring theme in the literature regarding negotiated settlements. It largely overlaps with the arguments that are being made that a negotiated agreement is beneficial for stability and democracy, because it incorporates both sides of the dispute. Joshi has ran an event history analyses for different civil war outcomes and found out that a democratic transition is more likely following a negotiated settlement than a military victory. He also stresses the importance of a balance of power within a country (Joshi, 2010). This makes sense, since a military victory by the government is likely to lead to a continuation of the status quo and the rebels will be unlikely to share power with their former enemies after a war if they win (Ibid.). He argues that rebels can only win by destroying the institutional capacity of a state and that this is critical for democracy (Joshi, 2010).

Toft, on the other hand, disagrees and shows that the recurrence rate for negotiated

settlements is twice as high as military victory and that their authoritarian score over time is also higher (2010). While she does not measure democratization, like Joshi, she does research the recurrence of large scale violence, which rarely ever occurs in a democracy. Also, both studies use a similar time frame and definition, so their contrary findings are remarkable. We have looked extensively at both studies in order to come up with an explanation for their opposite findings, but have not found a definitive answer. At least a part of the discrepancy lies in the fact that Toft distinguishes between negotiated settlements and stalemates, whereas Joshi puts these categories together. Toft further argues that negotiated settlements aim at preserving both sides as political actors. This tendency is detrimental for peace, because both sides have to compromise to a certain extend and thereby keep their grievances (Toft, 2010). Negotiated settlements also fail at providing assurances for both sides that their former enemies will respect the treaty (Toft, 2010). A one-sided military victory, on the other hand, will destroy the organizational structure of one side, which makes a resumption of the violence much more difficult (Licklider, 1995). Licklider finds that only 15 percent of military victories saw a renewal of violence within five years, where this was 50 percent for negotiated settlements (1995). Another argument is that a negotiated settlement will never

(11)

truly address a country’s problems, since the vested interests are still taken into account (Licklider, 1995). A military victory might alter this. When it comes to military victories, rebels were more stable in post-war situations than government victories, possibly because they have a certain legitimacy that is derived from the combination of popular support and dissatisfaction with the reaction from the government (Toft, 2010). They will also be more inclined to democratize, since rebels will generally be looking to change the status quo, whereas a government victory would often mean a continuation of the pre-war situation. Also, the government’s military is left intact when they win, which means they can use it to repress the population at large (Toft, 2010). This argument is supported by their data, which shows that governments that win civil wars remain about as authoritarian as they were before the conflict (Ibid.). This confirms that we have to differentiate between government and rebel victories within our category of military victories. Our analyses of the literature also shows that there is no consensus among scholars on whether a peace agreement or a military victory is beneficial for the chances of peace, which will make our findings more relevant. Both causal mechanisms that were described seem plausible and both have completely opposite findings regarding the best way to end a civil war. These large discrepancies in the

consequences of different civil war outcomes are a large part of the reason we wanted to further study the relationship between civil war and democracy. This study will further dive into the long term effects of both outcomes in order to see whether one of them makes democratic breakdown more likely.

2.3 Civil wars and democratic breakdown

The transformative nature of a civil war cannot be emphasized enough when studying it. This does not necessarily mean a change for the worse. Przeworski and Limongi even claim that many countries have democratized because of a civil war (1997). Given this starting point, it is not hard to imagine that a civil war might also influence the way that democracies

disappear. The cleavages that caused a civil war can endure, regardless of what political system a country has. Toft has argued that we need to extend the time scope of judging the success or failure of post-civil war environments, because our current standards are only focused on assuring a quick ending of violence and assume that combatants are generally willing to compromise in order to escape the constant fear of death (2010). These assumptions lead to a large emphasis on halting violence in peace negotiations, while it does not

necessarily address the underlying causes of the conflict. The focus on quickly stopping the violence might even be harmful, as it prevents conflicts from burning themselves out and could thereby serve as a welcome seize fire during which both sides can reconstitute and rearm themselves (Luttwak, 1999). The assumption that combatants will be willing to

compromise might also be naïve, as rebels often compensate their lack of resources compared to the government with intense commitment (Licklider, 1995). Such a rebel mentality does not accommodate moderateness. Also, their animosity against the government is what gives them legitimacy in the first place (Toft, 2010). Compromising themselves by reaching an agreement could therefore even be counter-productive to their support levels. Given that we possibly use a timeframe that is too short to judge the success of a civil war outcome, we might be overlooking democratic breakdown that occurs after our timeframe has passed. Toft shows that while most negotiated settlements between 1940 and 2002 were successful in

(12)

halting the short term violence, both peace and democratization ended quickly once she expanded the timeframe from five to ten or even twenty years (Ibid.). A focus on reaching compromise in civil war, as opposed to ‘give war a chance’ might therefore increase democratization in the short run, but also democratic breakdown in the long run. Another point of discussion is the extent to which having been a democracy before the outbreak of the war is associated with post-war democratization. Fortna and Huang have shown that there is an inverse relationship here (2008). Others disagree in their analyses and argue that previous experience with democracy helps to create durable settlements, which can help later (re)democratization (Hartzell, Hoddie & Rothchild, 2001). This large number of disagreements among scholars goes to show that this area of study is still developing and highly conditional, which is why more research is required. New studies continually disagree with existing literature, like the different patterns for pre- and post-cold war cases (Fortna & Huang, 2008).

A final point of discussion is the question of when to hold elections. This is the part where democratization meets democratic breakdown, because the first election could somehow undermine the results from the peace agreement, thereby making it a vicious circle. Where a civil war agreement carefully encompasses both sides of the conflict, certain democratic processes are able to undermine this through mechanisms like ‘the winner takes all’. A presidential system, for instance, precludes any possibility of power sharing, as the election can only be won by one candidate (Reilly, 2002). Joshi notes that elections could hurt the efforts to achieve political stability if former protagonists focus on their own stakes in the election outcome (2010). On the other hand, delaying elections can also make both parties more suspicious of the other (Ibid.). Former parties to the war who realize that they are unlikely to win an election may resort back to violence. As we mentioned earlier, Maeda has researched the extent to which different political systems affect external democratic

breakdown and found that this variable has no influence (2010). However, his research does not address internal breakdown, such as democratic leaders who refuse to give up power. Toft has already explained that the end of the twentieth century marked a steep increase in peace agreements (2010). Reilly argues that during this period the ’obsession’ with elections as a form of conflict resolution was at its peak (2002). If their expectations are true, we would especially see a lot of democratic breakdown after this period, because it has been associated with an increase of peace agreements and quick elections.

2.4 Mechanisms & application.

Each of the authors we have discussed above highlights a certain mechanism that can explain the occurrence of democratization after a civil war. What they all have in common is the notion that the ending of a civil war is an event that can create a unique dynamic through which democracy can spring or disappear. The main question seems to be how certain groups that have killed each other with considerable enthusiasm in the past now sustain a peaceful democracy, and why other groups can’t sustain this. When looking at the literature we discussed earlier, it seems clear that the occurrence of a civil war can be a factor in whether a country is able to sustain democracy, most likely for the worse. This makes us suspect that

(13)

democracies with a civil war in their recent past are more likely to break down democratically compared to those that don’t. This leads us to the following hypothesis:

H1: Democratic breakdown occurs more often in states that have democratized after a civil war compared to those who have not endured a civil war.

One of the more fundamental values that are important in the process towards a stable democracy, at least when it comes to a peace agreement, is some type of tolerance for the other. When such an accord has not been present, it seems like a different mechanism is at play, because the intentions of a rebel party that has been victorious are different. The rebels do not have to merge their preferences with another party, because they have defeated them. Their mechanism for democracy seems to stem more from a fundamental desire or ‘a normative preference for democracy’, as Mainwaring and Peréz-Liñán call it, to drastically change the situation in a country (2013). One could therefore expect them to be more successful in sustaining democracy, since there is fewer need to concede certain interests. Government victories that democratize are harder to explain. The fact that they democratize less often, as Toft suggests, seems logical since they fought to keep the status quo (Toft, 2010). In cases that government victories did lead to democratization, it may be out of sincere motivations or in order to prevent future uprisings. One would still expect democratization to occur less often and democratic breakdown more often. The different civil war outcomes and their prospective chances for later democratic breakdown will be an important part of this research. We will study them in order to see if the mechanisms established by the authors described above will hold in our event history analyses.

Our expectation is that peace agreements will have a lower chance of democratic breakdown, when compared to other civil war settlements. We thereby share ourselves on the side of Joshi, Mason and Wantchekon, mostly because their argument that only a peace treaty can incorporate both sides of the conflict in order for them to see an outcome that is preferable to a continuation of violence seems convincing. We also agree with the argument made by Joshi that a military triumph by either side does not incentivize the victor to take the others’

preferences into account. This would likely lead to new grievances, which over time could result in a renewal of violence and with it, democratic decay. In order for our expectations to hold true, we would need to see a lower democratic breakdown for states whose civil war was ended through a peace treaty compared to those that ended with a rebel or a government victory. These expectations result in the following hypothesis:

H2: the type of civil war ending affects the likelihood of a democratic breakdown.

This hypothesis allows us to research the extent to which different civil war outcomes have different chances for democratic survival. It also means that we can still accept our hypothesis if one of the other civil war endings turns out to have an impact on democratic survival.

2.5 Control variables

While studying democratic breakdown, it is important to take factors into account that could interfere with our results. An important way to do this is by using control variables, who make sure that there are no other factors which can explain our phenomenon. Like other studies into

(14)

democratic breakdown, we will therefore include a range of control variables to solidify our results. We will explain why we have chosen to include each of our control variables below. The first set of covariates that we will include are those that relate to the civil war itself. Firstly, we will control for the civil war dummy in order to see if a history of a civil war in itself increases the likelihood of democratic breakdown. It could be the case that this has an intervening effect on our findings, because having had a civil war already affects democratic survival. Due to this possibility, we will control for the civil war dummy in a separate analysis. Secondly, we will control for the duration of the civil war. We have just discussed the theoretical debate over this covariate. We will include it to see if there is evidence for the idea of ‘war fatigue’.

Besides civil war covariates, we will also control for a factor about which there is a general consensus that it matters to democratic survival: the economy. Even though some authors don’t agree for certain regions, there is general consensus about the idea that economic factors can have a significant effect on the chances for democratic breakdown (Joshi, 2010;

Przeworski & Limongi, 1997; Boix and Stokes, 2003). However, Przeworski and Limongi have found that economic growth is more important than other economic variables, as ‘’a democracy is more likely to survive in a growing economy with less than $1,000 per capita income than in a country with an income between $1,000 and 2,000 that declines

economically’’ (1997). This means that it is not the level of development, but the current economic climate that we are interested in. We have therefore chosen to control for economic growth.

Also, we have to take natural resources into consideration. There has been extensive debate regarding the ‘resource curse’. Most of the academic studies into the causal mechanism between oil and democracy indicate that the presence of oil is detrimental for the

democratization process (Ross, 2001). However, it remains unclear if oil also play a role in the democratic breakdown of states, since there has been very little research into the matter. We will therefore include the annual production of crude oil, coal and natural gas to our analysis as covariates.

We have also discussed the extensive debate regarding the political system of a country and its effect on democratic decline. The debate focused on the differences between presidential and parliamentary systems of government and how these differences could result in a higher number of democratic breakdowns. Maeda has argued that presidential systems have a higher chance of breakdown in this regard. We will test if his findings hold in our analysis.

Lastly, we will control for the number of democratic breakdowns that a country has already encountered. There has been very little debate about the consequences of a previous

democratic episode. Fortna and Huang have argued that having been a democracy before the outbreak of a civil war may lead to citizens viewing their democratic political institutions as a failure (2008). Others have argued the exact opposite, claiming that countries with previous democratic regimes are more used to accommodating competing interests which is beneficial for peace and democracy (Hartzell, Hoddie & Rothchild, 2001). The effect of previous breakdowns has not been included in previous studies. This makes it hard to predict the consequences.

(15)

Chapter 3: Methodology

This chapter aims to explain the research methods, the operationalization and the variables that are used for the analyses which is performed in Chapter 4. We will start off by laying out some of the more complicated aspects of the research method, like what our choice for the Cox regression entails for our findings. The chapter will proceed with the operationalization of the key concepts of this thesis. Operationalization is important in order to effectively demarcate the boundaries of what is being researched. The chapter finishes off by describing the different control variables that we will test.

3.1 Research method

Our general goal in this research is to investigate what influence different characteristics and conditions of a country can have on the life expectancy of its democracy. More specifically, we attempt to determine what the influence of different previous civil war outcomes is on democratic survival, while controlling for a number of other influences. A survival analysis is a well-suited way to study this effect, because it specializes in determining the duration until the occurrence of an event. At the same time, it is able to take other control variables into account.

The term ‘survival analysis’ is an umbrella term for a number of survival functions, all of whom have different characteristics. These characteristics make each survival function suitable for different types of research. When it comes to our study, the main difficulty is that we have a large number of different independent variables that influence the survival time (also known as covariates). Non-parametric tests like the Kaplan-Meier estimates do not allow for multiple covariates, which is also the reason we won’t use this method. This leaves us with the parametric and semi parametric tests. What makes it even more difficult is that the covariates that we intent to use include both time-varying and fixed covariates. These terms are used to describe whether a covariate changes over time or whether it normally stays the same for a country throughout time. An example of a time varying covariate would be a country’s gross domestic product, whereas its political system would be fixed. Another important aspect of choosing a suitable survival analysis is the assumptions that we can make about the hazard function. Semi-parametric tests try to establish the relationship between the covariates and the hazard of the event occurring, whereas parametric tests already roughly assume what this relationship looks like. This means that having to make concrete

assumptions about the way civil war influences democratic breakdown is the main difference between our two remaining functions. If we were to use parametric tests we would already need to know what the relationship looks like and use this knowledge to gain a clearer picture of the precise details of the relationship (Mills, 2011). However, the exact direction of this relationship is exactly what we want to examine in our research. This makes parametric tests unsuitable for our research. That leaves us with semi-parametric tests, which make no assumptions about the relationship, but do allow us to use multiple covariates (Mills, 2011). The best known semi-parametric test is the Cox regression model. One of its characteristics that we need to discuss is the proportional hazard assumption. This assumption entails that any hazard of a particular case is a fixed proportion of the hazard of any other case (Mills, 2011). This means that the ratio of the hazard is constant over time. In other words, the log

(16)

hazards of different cases are parallel to each other. Since this is a core feature of any Cox regression, we have tested whether it holds true for our data. We have included the

proportional hazard tests in appendix 4. As we can see under the P value, only the system covariate is significant for this test. This entails that we cannot run a cox proportional hazard analysis while controlling for only our system covariate. The system covariate no longer has a significant P value if we include it in an analysis with other covariates (see table 7.4.2). We still need to keep the subject of democratic breakdown in mind when thinking about different research methods. It is clear that not every democracy has a similar hazard of breaking down, even when we take our covariates into account. Some countries are just more prone to democratic breakdown. One possible reason for this discrepancy is that they have experienced it before. The idea that countries with a case of democratic breakdown are more prone to a new breakdown seems logical, since there has already been a successful attempt to undermine the democratic institutions. The fact that a democratic breakdown has already occurred suddenly makes it a viable and realistic threat for the future. One could even say that the political elite has a roadmap of how to orchestrate a take-over, since many of the

cleavages have often remained. The same could be said for a country’s population. Ifidon notes how collective memories of past ‘wrongs’ could trigger a number of events that lead to democratic breakdown, either intentional or not (2002). Democratic breakdown can therefore be considered as a ‘recurrent event’. This means that we observe a correlation between similar events within a single object (Mills, 2011). Also, because the risk of a recurrent democratic breakdown applies to all countries and can even occur multiple times, we consider it a ‘shared frailty’. Once a country has experienced a democratic breakdown it becomes more fragile than others and we need to account for this effect. We will do so by introducing a frailty model in our analysis. A frailty model aims to account for the excess risk for distinct subgroups within our dataset (Mills, 2011). Including a frailty model is a relatively new development and has become more or less standard in new survival analyses (Ibid.). In our case, the subgroups that we will account for will be cases that have had one or multiple democratic breakdowns within our time frame. Our eventual analysis will show if previous breakdowns indeed hurt future attempts of democratization. However, it is worth noting how little has been written about this subject. Almost no study of democratic breakdown takes frailty into account. We have also found very few studies that attempt to explain the mechanisms behind a recurrent democratic breakdown.

3.2 Operationalization

When discussing our survival analysis, it is important to clarify a number of conditions related to the workings of our method. We will also further clarify and limit the terms we use below. We will start off by explaining what we consider as democracies and at what point exactly they break down. Afterwards, we will do the same for a civil war. Finally, we will explain our control variables and clarify how we measure them.

3.2.1 Dependent variable

We have already explained how most scholars of democratic breakdown use a low bar for measuring democracy, because it is easier to detect differences that way. We follow Dahl’s 1971 classic ‘Polyarchy’ in treating all regimes that hold elections in which the opposition has

(17)

some chance of winning and taking office as democracies (Dahl, 1971). Having elections in which changing the government is a possibility avoids violence, because they maintain a prospect for long term retrieval of power (Przeworski, 1999). While violence might represent short term benefits to the losing side, it does not match up to the long term benefits of

possibly regaining office. The measurement of democracy that we use from the Polity IV dataset is the variable called ‘Polity2’. It is the democracy score (DEMOC), subtracted with the autocracy variable of Polity IV. This makes our variable more robust than the democracy score, because it filters out regimes that have a positive score on one of the indicators, but should have a negative score due to deeply authoritarian scores on the other indicators1_{. An}

example of such a case would be a country that has competitiveness, but no openness of executive recruitment. This makes the positive democracy score meaningless, since it heavily limits who can join the opposition. There are numerous countries that display such scores. Iran, for example, has fair and competitive elections. However, all of the candidates are selected and approved by the Guardian Council, which is led by the Ayatollah (Al Jazeera, 2013). Another feature of the Polity2 variable is that it records changes in cases that are in transition by the average change per year. This is beneficial because it shows us the entire period during which a country transforms. The variable of democracy and autocracy that are used are derived from the competitiveness of political participation, the openness of executive recruitment and the constraints on the chief executive (Marshall, 2016). All are weighed equally, they are just measured reversely for democracy and autocracy. As Fortna and Huang indicate, most cases of democratization are autocracies that liberalize, but do not fulfill the full definition of a democracy (2008). This is likely to be similar, but reversed for democratic breakdown. This makes it useful for us to use an indicator that also signals the decline of autocratic characteristics. Minor steps in either direction could be important, because descending below a certain democratic standard could clear the way for more repressive methods. Another benefit of our approach is that it allows us to look at cases of democratic breakdown, even when they are not necessarily accompanied by regime change. As Alemán and Yang argue, regime changes are often incremental and long (2011). Focusing too much on actual regime change would then weaken our research, because we would be ignoring a lot of cases.

Since we only use two different regimes in our eventual categorization, democracies emerge when autocracies disappear and vice versa2_{. The middle area of contentious regimes are}

normally called ‘anocracies’. Polity classifies anything with a score from +1 up to +5 as ‘open’ anocracies and anything from -5 to 0 as ‘closed’ anocracies (Ibid.). This distinction is crucial, as most countries that were previously war-torn go through a decisive phase where they are either an ‘open’ or a ‘closed’ anocracy before ending up as fully democratic or autocratic (Cole and Marshall, 2014). This finding is crucial, since it means that a lot of countries won’t bounce back from democratization of democratic breakdown towards the other end of the spectrum. The evidence for their findings was visible after World War II,

1 This can become a problem because the democracy variable that is used is an additive variable, where any positive indicators can lead to a positive overall score of democracy.

2 The democracy and autocracy variables are eleven point scales, measured from 0 to 10. The Polity2 scale that we use is a single regime score that ranges from +10 (full democracy) to -10 (full autocracy). Within this 21 point scale, Polity distinguishes between autocracies (-10 up to -6), anocracies (-5 up to +5) and democracies (+5 up to +10) (Cole and

(18)

during which period an unusually high number of anocracies existed (Cole and Marshall, 2014). This makes sense, since the devastation of war can hollow out political systems, leaving them vulnerable for transition of any kind. Over time, these anocracies were solidified into either democracies or autocracies (Ibid.).

Open anocracies are not fully democratic, but still resemble more democratic than autocratic features. We therefore consider any state with a score higher than 0 as democratic and any state with a score of 0 or lower as autocratic. This entails that any state that had a Polity2 score of higher than 0 in one year and a score of 0 or lower the following year is considered as a case of democratic breakdown. Democracies breaking down and then emerging again will be different episodes and therefore different cases. The reason we will use a low standard of democracy is in order to have a more sensitive measure of when democracies are in decline. As we explained earlier, the cases of democratic breakdown will most likely be the young, unstable democracies. This is due to a number of reasons. Firstly, young democracies make it easier for a single person to gain power because of less effective checks and balances.

Secondly, unstable democracies render insurgency more beneficial due to weak local policing and often through corruption (Fearon & Laitin, 2003).

3.2.2 Independent Variable

When it comes to our independent variable, the civil war score, we will also make use of the Polity IV dataset. In this dataset, CivWar (civil war) is measured by the number of major episodes of political violence (Mepv) that consist of at least 500 directly related deaths over the course of an episode (Marshall, 2017). Each episode is measured in years. The difference with other measurements is that this definition does not necessitate a certain number of deaths per year. A broader definition also entails more cases, which is beneficial for our analyses. It also allows us to study civil wars that have continued for longer periods of time, but have not had a high intensity throughout the conflict. This makes the dataset more suitable to research the claim by Gurses and Mason that conflicts which have lasted longer create a war fatigue and thereby result in higher chances for democratization (2008). A country is scored between 0 and 10, where 0 denotes no episodes of civil war and 10 denotes the highest on a ratio scale (Marshall, 2017). A higher number might also be a civil war which has had several episodes of violence within a single year. For instance, if a country sees two level 3 civil warfare episodes, this is categorized in the same way as one level 6 episode. Multiple level 1 or 2 episodes within a year can also account for a level 3 episode. We use a level three civil war as threshold for our research, because it is clearly more significant than protests or riots and the acts are not confined to short periods of time, as is the case in category 1 and 2. Instead, category 3 entails objectives that are focused on strategic authority, possibly long period of relative quiescence and a death toll between 10 and fifty thousand per episode (Marshall, 2017). There are 18 separate civil wars that add up to a level 3 score or higher. These civil wars have all occurred in a different country.

Furthermore, we have distinguished two separate variables regarding civil war that we will control for. These are the civil war outcome and the duration of the civil war. We have coded each of these variables in a different way. When it comes to the civil war ending, there are four options: a peace treaty (coded as 1) , a rebel victory (2) or a government victory (3). We

(19)

have eliminated any ongoing civil wars from our dataset. Ongoing civil wars are redundant for our analysis, since they have not produced a civil war outcome yet. The civil war duration was calculated as the number of years that the civil war has lasted. We have measured from the moment violence erupted or war was declared up until one of the outcomes mentioned above was reached.

3.3 Control variables

In the theoretical section, we already explained why we use the control variables that we include. In the text below, we will explain how we will use them. Some of these control variables are time-varying, which means that their score differs from year to year (economic growth, oil production, etcetera). Others are time-invariant, which means that their score normally does not change over time (political system, geological location).

The first set of control variables we discussed are those related to civil war. The first civil war covariate is the occurrence of a civil war. As we mentioned during the theoretical chapter, this will test the degree of intervening effects of a civil war itself when we are measuring the effects of the different civil war outcomes. We will also control for the duration of a civil war. We have discussed the idea of ‘war fatigue’ during our theoretical chapter and how this can affect the likelihood of peace. If this concept holds, we would see increased democratic survival rates the longer a civil war has lasted.

To get a clearer picture of the contemporary economic climate that a country is in, we have adjusted our GDP values into the annual economic growth of the GDP per capita variable. We are talking about a lagged variable, as it is a direct derivative of the GDP per capita variable. This implies that the data here starts one year after the GDP variable for each case, as we can only calculate growth if we have a second value. Our expectation with the economic growth (EG) covariate is that countries will break down sooner if economic growth is lower. We have found data from the World Bank regarding every country’s GDP, starting from 1950 for most countries and in 1960 for some other countries (World Bank, 2018).

We will also control for oil, natural gas and coal by using control variables regarding the annual production of each of these in a specific country, measured in annual Megawatt of production per capita. We will use coal, oil and gas as three separate control variables, because their respective influence is likely to differ greatly. Where the price and usage of coal has diminished over time, oil has become more and more important for both energy purposes and for producing goods (Radetzki, 1995). Merging these variables could therefore give a wrong picture of the importance of natural resources to a state in a certain period.

In our theoretical chapter, we have already explained Maeda’s finding that presidents are more likely to become authoritarian than prime-ministers (2010). It will thus be interesting to see if our data supports this finding. We do so by controlling for a country’s political system. In order to use this variable, we will use Lijpharts’ measurement for executive-legislative relations (2012). We have coded this covariate as follows: 0 is a parliamentary or an assembly elected president and 1 is a presidential system. If Maeda’s findings are correct, we would see an increasing correlation with democratic breakdown the more presidential a system becomes.

(20)

As we have mentioned, we will count the number of democratic breakdowns that a particular country has had in order to accurately measure the control variables. Previous failed attempts to deploy democracy might increase the likelihood of democratic breakdown. We will use a variable (Frailty1) in order to indicate the number of times that a country has fallen below the democracy threshold after surpassing it. More specifically, this means that we will count the number of times that a country’s score has dropped below a Polity2 score of 1, after initially reaching 1 or higher. Adding a shared frailty mode has become a standard approach in event history analyses (Mill, 2011). However, other survival analyses into democratic breakdown do not include such a frailty score (Joshi, 2010 & Maeda, 2010). This makes the results of our frailty score even more relevant.

3.4 Data

Our data is structured as separate countries. Each country is then divided into the number of democratic years. If a country’s polity score reaches below zero, the country temporarily disappears from our dataset until it regains a democratic score. Each year is a separate case. Our eventual dataset encompasses a total of 4606 observations over 142 countries, each of whom has been democratic at some point between 1945 and 2015. The duration of this period ranges between a couple of years for countries like Yemen and the entire 71 years in countries like Canada. When using the definition of democracy and democratic breakdown that we have laid out in this chapter, we find 99 separate cases of democratic breakdowns during this period (see below).

(21)

Variable Valid N Min Max Mean /

Percentage Std. dev.

Dependent variable

Democratic breakdown (Dummy) 0 1 4606 4507 99 0 1 97,9% 2,1%

Civil war covariates

Civil war occurrence (Dummy) 0 1 4606 4588 18 0 1 99,6% ,4%

Civil war duration 18 3 16 8,9

Economic covariates

Economic growth 4017 -,540 ,937 ,0295 ,0630

Natural resource covariates

Coal 2348 ,000 149,5 6,400

Crude oil 2963 ,000 417,8 9,971

Natural gas 2087 ,000 239,6 8,119

Institutional covariates

Political system (Dummy) (Semi) parliamentary (0) Presidential (1) 3491 1932 1559 55,3% 44,7% Frailty

Cumulative number of democratic breakdowns (Frailty) 0 1 2 3 4 5 4606 3426 792 276 71 40 1 0 5 74,4% 17,2% 6,0% 1,5% 0,9% 0,0% Table 3.1: descriptive statistic of the dependent variable

(22)

3.5 Outlier analysis

We have checked our results for the influence of outliers with the use of the DFbeta function in R. This function shows us how much our coefficient would change if R were to take out the most influential case in the analysis. It is important for us to run this test, because we could find false values or values that are really extraordinary in a certain way and therefore corrupt our findings. The formula 2/sqr T(number of observations) shows us how much any

influential case is normally allowed to change our coefficients. In our dataset this formula is equal to 0.03155579. This number is based on the assumption that each observation contains a value for the dependent variable. However, our analysis contains only 99 separate democratic breakdowns. This low number of observations is inevitably going to lead to large outliers, simply because individual observations carry more weight in a smaller sampling size. We have included the results for each of the individual DF beta functions in our appendix. When looking at our civil war ending dummies, it becomes clear that we have one or two outliers per covariate that seem to dominate all of our findings. However, none of these outliers seem like obvious candidates to be influential cases for our civil war endings. Mostly because none of them have even experienced a civil war. If we remove these countries from our dataset and run the DF beta analysis again, we end up with different outliers from a similar proportion in the same place on the plot. The reason we keep getting these outliers at the same place in our outlier plot is that we are dealing with rare events. The plot for our economic growth covariate is different (see appendix 7.3.3). This covariate has far more observations compared to the civil war endings. With more values come more outliers. However, removing these outliers does not change the results of the economic growth covariate in any way. This suggests that the outliers are influential, but not divergent in the sense that they are not far removed from the curve for the economic growth coefficient.

(23)

Chapter 4: Empirical analysis

In this section, we will analyze the results from R studio in order to determine to what extent our hypotheses hold true. We will first discuss the basic survival function, which shows the general chances of a democracy’s survival rate. Afterwards, we will see to what extent previous civil war outcomes matter to democratic survival. We will then control for the other covariates that we have discussed. Finally, we will check the robustness of our results by doing an outlier analysis.

4.1 Basic survival function

When we run our initial Cox Proportional Hazards function without controlling for any of the covariates we get out basic survival function. This plot gives us a preliminary idea of what the survival odds for democracies look like. We have included the plot below. On the X axis the plot shows the number of consecutive years that a democracy has existed. This starts in the year 1945 for most countries, but it also includes countries who were founded after 1945. If a democracy were to break down and later regain a positive democracy score, it also starts over from the year ‘0’. On the Y axis, the plot shows the probability that a democracy survives for the given number of years.

Plot 4.1: basic survival function.

One of the first things we notice about this survival curve is that it flats out. This indicates that young democracies are fragile, whereas the chances of democratic breakdown become very slim once a democracy has survived for a longer period of time. This supports the argument made by the Polity scholars that regimes go through a contentious early phase before stabilizing (Marshall, 2017). Our basic survival function shows that this early phase of a democracy lasts about thirty years and that this point marks some kind of threshold. After this the chances of a democratic breakdown are really slim. Some scholars have referred to this phenomenon as ‘consolidating’ (Svolik, 2008 & Diamond, 1999). The only exception is the case of Venezuela, whose democratic system broke down after a period of 53 consecutive

(24)

years, when Hugo Chavez came to power. Polity’s claim that the early phase of democracy’s is contentious is further supported by the fact that half of democracy’s that break down do so in the first ten years. This goes to show how fragile young democracies really are, even despite the institutional support they often receive. Svolik has similarly claimed that the existing models underestimate the risk of an early reversal, while overestimating the risk of a late reversal of a democracy (2008).

4.2 Adding the civil war covariate

Before looking at the different civil war endings and their influence, it is important to have an idea of just how influential a civil war itself is to democratic breakdown. We have added the table for our civil war dummy below. This goes to show that having had a civil war has a statistically significant impact on democratic survival rates. Furthermore, the effect that it has is quite significant with an exponential coefficient of .4193, which means that having endured a civil war before decreases the odds of democratic breakdown after a civil war is about 42 percent of the chance that other countries face. This finding is quite remarkable. Judging from the civil war literature about its long term effects, we would expect that civil wars have a negative effect on democratic survival. Instead, it seems like the opposite is true. This finding also has implications for the results of our analysis into the different civil war outcomes. It means that we have to take this finding into account, because it could be an intervening variable to our civil war ending findings.

Table 4.1: Cox proportional hazards analysis while controlling for the civil war dummy.

Coef Exp (Coef) Exp (-Coef) P value

Civil war dummy -0.8692 0.4193 2.385 0.0140

Frailty 0.0008

Once we add a number of control variables to our analysis, it becomes clear that the statistical effect of the civil war dummy vanishes (see below). We have previously stated our hypothesis as follows:

H1: Democratic breakdown occurs more often in states that have democratized after a civil war compared to those who have not endured a civil war.

The effect of the civil war dummy has a negative value in both of the analyses. This means that, in both cases, having endured a civil war seems to aid democratic survival. This alone is enough to reject our first hypothesis, since it predicts that survival rates will be lower for countries with a civil war. The table below is also clear that there is no longer a statistically significant difference between the democratic survival rates of both groups once we add some control variables. This is another reason to reject our first hypothesis.

We have also included the duration of the civil war in our analysis to see if there was evidence for the idea of ‘war fatigue’. We have included the table in our appendix (see table 7.1.3). The results clearly show that this covariate is both statistically insignificant and without any relevant effect. This means that civil wars that have endured over longer periods of time do not yield more stable democracies.

(25)

Table 4.2: Cox proportional hazards analysis for the civil war dummy with additional control variables.

Coef Exp (Coef) Exp (-Coef) P value

Civil war dummy -0.6392 0.5277 1.895 0.1900

Political system -0.2706 0.7629 1.311 0.3900

Economic growth -7.2525 0.0007 1411.7 0.0019

Frailty 0.0026

4.3 Controlling for civil war effects

In the table below, we are looking at the results that different civil war outcomes have on a democracies’ chances for survival. The government victory dummy is not visible, because we have taken it as the reference category for our dummy’s. We can see that only one of the dummy variables is statistically significant, the peace treaty dummy (PTD). The rebel victory dummy (RVD) is not statistically significant with a 5 percent certainty margin. The score of our peace treaty dummy hazard ratio (exp (coef)) means that the chances of democratic breakdown are 24 percent of those with a different civil war ending. So countries that

democratize after a civil war that ended with a peace treaty are about four times less likely to break down compared to countries with a government victory. This finding is quite

remarkable, because it suggests that there are significant differences between civil war

endings in terms of their effect on democratic survival rates. The table above indicates that the peace treaty dummy is not just the only dummy that is statistically significant, but also has a positive effect on democratic survival. This finding is in line with the findings of some scientists that peace treaties are beneficial to long term peace. It is also a preliminary answer to our main research question. It seems to suggest that we can accept our hypothesis, if this finding holds when we add our control variables. However, the findings also run completely contrary to the arguments made by Toft and Luttwak, for example, who have claimed that peace treaties often hurt long term stability.

Table 4.3: Cox proportional hazards analysis for the civil war outcomes

variable coef Exp (coef) Exp (-coef) P value

Rebel victory dummy -0.375 0.6874 1.455 0.6000

Peace treaty dummy -1.440 0.2370 4.219 0.0460

Frailty 0.00082

What also seems to discard Toft’s theory is the fact that the rebel victory dummy does not have a P value below 0.05. This implies that there is no relationship between a rebel victory and democratic survival. It also seems to discard the idea of ‘give war a chance’, since both the rebel and government victories don’t seem to aid democratic survival rates.

We can also see that the frailty score here is quite high. This goes to show how great the impact of a democratic breakdown is on state fragility. It seems that failed attempts to democratize carry enormous consequences for future attempts with them. It is surprising to see how little has been written about this phenomenon, once we look at the massive impact that it seems to be having. One important point is that these results are still without controlling for the covariates, which we will do below.

Out of the frying pan into the fire?