Why does civil war recur? Civil wars are more frequent, more deadly, and last longer than wars between nations (Walter and Snyder 1999; Frearon and Laitin 2003). In a broader sense, civil wars are known to create the “trap of conflict,” where conflict undermines the foundation of the state, and the state becomes vulnerable to another conflict. Once a state becomes trapped in it, the state is exposed to vulnerabilities and can thus be caught in another trap. Struggling to escape the trap leads to a lot of bloodshed and leaves stains on the trap. This blood includes the blood of innocent civilians (noncombatants) and becomes part of the stains on the trap.

After a certain period of peace, many states tend to fall back into the trap of conflict. Approximately 75% of states experience conflict, based on data from UCDP intrastate war version 21.1 (see Appendix 1 for a detailed list of civil war onset) and the estimation of civil war recurrence is approximately 35% for the next decade. Limiting the data to the same protagonists (belligerent), 36 out of 97 states were trapped back into conflict with the same rebels. Ethiopia, Angola, India, and Myanmar, for example, experienced more than 15 episodes of civil wars in total, and on average, most conflicts started one year after the previous civil war. However, France and Greece did not experience any civil war after the civil wars in the 1950s and 1960s. What makes this difference? What factors contribute towards countries to avoid these traps of conflict while others fall back into it?

Many scholars who examined the recurrence of civil war have noted that conflict likely recurs for the same reasons that civil wars begin—for example, poor economic development (Fearon and Laitin 2003), third-party intervention (Walter and Snyder 1999), a low-level of democracy (Barnett, Fang and Zürcher 2014), and ethnic fragmentation (Gurr 2000), but these reasons are not convincing enough because they do not consider the effect of civilian casualties in civil war on the durability of peace.

This research posits that attacks on civilians during previous civil wars continue to affect the failing peace (difference in ideologies) after the civil war for the following reasons. First, the indiscriminate violence against civilians during the conflict serves as a mechanism for civilians to join the opposition and increases their willingness to resist (Salyvas 2006). Second, when the civil war ends, both sides of the combatants or noncombatants associated by supporting either government or insurgency, are trapped in the same territory with the same people who earlier tried to kill them. In contrast, after war between nations, combatants return to their homeland. Due to these fundamental differences, it is more difficult to achieve long-lasting peace after the civil war, rather than in wars between the nations (Licklider 1995).

Given these perspectives, however, studies paid less attention to innocent victims, and did not fully address the dynamics of civilian victims. Therefore, the current study quantitatively analyzes 407 civil war cases from 1946 to 2021 to determine the causes of recurrence and examines how civilian casualties affect the post-civil war peace duration. Based on the results of the empirical analysis, it is possible to derive a detailed understanding of the influence of civilian victimization (intentional use of violence) on the duration of post-civil war peace.

Killing Civilians as a Strategy?

Killing and victimizing civilians during the conflict poses a great risk for both governments and insurgents. Such actions can lead to the loss of support, as well as strengthening the adversary’s will to resist, which is interpreted as “self-defeating” and “counterproductive” (Kalyvas 2006). However, targeting (or victimizing) civilians is one of the strategies for coercing change in the adversary’s behavior or weakening military capacity. These strategies are considered methods of punishing the enemy government (in a conventional war) or rebel group (in a guerrilla war) in hopes that civilians will rise and demand the end of war. In this relation Valentino, Huth and Balch-Lindsay (2004) also noted that guerrilla warfare provides a strong incentive to kill large numbers of civilians, especially in conflicts with large numbers of insurgents and widespread support among the population.

This argument is that the motives of the two groups differ based on the population. Insurgents with broad support from the population can pose a threat to the government, and rebels long for and need extensive support from the population to counter the government. However, insurgents that are not supported by the general population are no longer a threat to the government. Therefore, the population plays a significant role in disputes. In particular, the government can find itself amidst a complex structure of conflict wherein it competes with various insurgent groups. Although insurgent groups may also compete with other insurgent groups in the complex structure of different demands and interests, when falling into such a complex trap, the general population plays a key role in escaping from the trap.

Many scholars of civil wars have noted that protecting civilians is a way to elicit cooperation and support (Trinquier 2006). Both governments and insurgents must minimize the use of violence against noncombatants. By doing so, they can maintain or strengthen their legitimacy and political will with the support of civilians. Nonetheless, use of violence shows remarkable variation in civil wars. TABLE 1 presents civilian casualties in civil wars.

TABLE 1 shows civilian casualties estimated by decade. From the 1980s to 2020, 1,813,850 civilians were killed by governments and rebels.² Among the overall number of casualties, 1,486,157 were noted on the government side, and 327,693 civilian casualties were noted on the rebel side. Data from the 2000s to the 2010s indicate that insurgents are responsible for killing more civilians during the conflict. The use of violence against civilians clearly arises and is used as a strategy for both governments and insurgents and it is an inevitable consequence of conflict. Scholars have agreed that violence against civilians is a “double-edged sword,” even for its perpetrators (Valentino 2014).

There are two primary reasons why this strategy is employed in civil wars. First, civil wars fundamentally differ from interstate wars. It is a matter of existence, where the parties generally continue to exist within the same borders after the end of the war. However, defeat can mean the loss of existence, and such high stakes typically make compromises difficult (Licklider 1995). In addition, these high stakes make actors desperate for victory and drives the use of all means and methods which consequently leads to mass killing (Downes 2008). Moreover, when attacking the opponents, it is hard to separate the general population from the opponents. In particular, the military base used by insurgents is typically located in the cooperating villages or control areas where noncombatants live.

Second, the complex dynamics differ from that of interstate wars, often because there may be more than one rebel group that the government must deal with in terms of domestic disputes (Jentzsch, Kalyvas and Schubiger 2015). For example, Syria currently hosts numerous competing insurgent groups, including the largely secular Free Syrian Army, local Islamist factions (such as the Syrian Islamic Liberation Front and the Syrian Islamist Front), Al-Qaeda-linked groups like Jabhat al Nusra, and many smaller factions. In Syria, all insurgents similarly seek to take over Bashar al-Assad’s government. Meanwhile, rebel groups commonly compete with one another by using violence to control territory, resources, and civilian support (Morris, Warrick and Mekhennet 2013). A recent study proved that violence against citizens increases when competition among rebel organizations intensifies (Wood and Kathman 2015).

High stakes and complexity of conflict further intensify desperation for victory in civil wars and eventually lead to numerous casualties. Research has claimed that desperation for victory in intrastate wars, especially in the war of attrition, can lead to states (or insurgent groups) taking part in barbaric massive killings (Downes 2008; Wayman and Tago 2009). State-sponsored mass killings are more likely to occur during irregular civil conflicts (Krcmaric 2017).³

Therefore, from this perspective, killing massive numbers of civilians as a strategy is more common during civil wars. To sum up, violence against civilians, regardless of whether it is perpetrated by the government or rebels, is used strategically, and even genocide is used as a strategy in civil wars. However, does this strategic choice really lead to long-lasting peace after the civil war? This question will be discussed in the next section.

Raising the Anger?

Every civil war ends with either a rebel victory or a government victory, as well as a negotiated peace settlement, or other similar aspects. After the termination of civil wars, civilians, including those involved in the civil war, are trapped in the same territory. So, when they were trying to kill each other until yesterday, can they cooperate and maintain peace? Is it possible to sleep with the enemy?

In recent years, scholars have argued that the risk of recurrence increases or decreases depending on the outcome of the civil war (Ohmura 2011). However, none of them reflected findings from literature using violence against civilians in earlier civil wars. Civilians who were exposed to violence, or their families and friends, should also be reflected in the study of durability in post-civil war peace. For instance, affected civilians, or friends and family members of victims of violence, will live in sorrow and blame the use of violence for hurting their loved ones even after the termination of the civil war (Lyall 2010).

There may be many reasons why states fall back into the trap of conflict after the termination of civil war, but using violence against civilians has not been considered as a variable for sustaining peace afterward. For example, family members or friends of those killed in conflict may want to take revenge on those who have taken the lives of their loved ones. Therefore, after the end of civil war, they may become supporters of rebels or other group members to take revenge on those who used violence. This violence against civilians can result in either strengthening their resolve or persuading them that they have nothing to lose by joining the opposition (Kalyvas 2006, 154; Valentino 2014).

For example, the civil war in Senegal, Casamance has proved to be disastrous for the local population. The Movement of Democratic Forces of Casamance (MFDC) demanded its independence in the Casamance region, whose population is religiously and ethnically distinct from the rest of Senegal. The Senegalese government was non-democratic, and the rebels were considered weaker than the state’s military force (Gleditsch, Cunningham and Salehyan 2013). The Senegalese government took advantage of this fact conducted air strikes and bombings of military bases in controlled areas of the MFDC.⁴ The Senegalese military persecution and attacks on their villages resulted in civilians joining the MFDC movement (Gehrold and Neu 2010). This phenomenon strengthened the MFDC movement and boosted their ability to resist attacks. According to the dataset from UCDP One-sided Violence, seven civil wars occurred from 1992 to 2001 and at least 226 civilians (non-battle-related) were killed in government attacks during this period. The Senegalese government's attacks led the country into a trap of conflict.

Consequently, the aforementioned case and theoretical discussions on targeting civilians can be interpreted as follows. Targeting civilians served to further embolden the insurgency act and led to extreme backlash because it gives civilians, who otherwise may have sat on the sidelines or even supported the government, a reason to cooperate and side with the guerrilla insurgents (Krcmaric 2017). Therefore, the following hypothesis is posited.

Hypothesis 1: Governments targeting civilians during the previous civil war will fail to maintain peace. Moreover, they will likely fall back into a trap of conflict.

Raising the Enemy?

In the previous section, a hypothesis was posited by citing theoretical approaches and cases regarding the potential consequences of the government's attack on civilians. This section presents cases and hypotheses of how civilian attacks affect insurgents.

Rebels are driven by regional, ethnic and religious motives, therefore their civilian support base is limited and weak, compared to governments (Stanton 2015). Rebels thus strive to gain widespread civic support. Insurgents particularly rely on assistance from civilians when operating in particular regions (Kalyvas 2006, 148). Moreover, they set up village-based administrations to resist the government while taking control of that area. In particular, civilians can help by providing shelter, food, direction and with collective information that allows them to address identification problems effectively (Wichham-Crowley 1990, 216-217). Rebels, receiving such a strong civilian support base, are bound to put pressure on the government. Furthermore, in this case, even if a civil war ends in government victory or peace treaty, the rebels can resist the government again and cause another civil war when they are unsatisfied.

Despite the need for widespread civic support, insurgents take the risk of killing civilians to resist the government and coerce demands. In this situation, civilians who have lost their hometown or were forced to leave the village due to insurgents are likely to go under the protection of the government. This eventually weakens the civilian support base in the long term and weakens military resources, such as supplies, resources, and weapons, for insurgency. In this regard, Condra and Shapiro (2012) claimed that when the insurgents target civilians, they seem to lose support from civilians. In particular, concerns about rebels losing civilian support were found in a letter to Iraq’s al-Qaeda leader Abu Musab al-Zarqawi in 2005; al-Qaeda's second leader, Ayman al-Zawahiri, urged him to minimize indiscriminate violence against Shiite civilians (Filkins 2006).

The events in Jama'atu Ahlis Sunna Lidda'awati wal-Jihad, otherwise known as Boko Haram in Nigeria, provide compelling evidence related to the above. On 15 April 2014, rebels operating in Nigeria made worldwide headlines when they abducted 276 Nigerian secondary school students.⁵ They acted with the aim of coercing the government through threats and gaining international notoriety for their violent campaign of terror (Bamidele 2014). They were criticized not only for abducting the girls, but also for their broad use of violence and terrorism. This kind of resort to violence has been a key subject in academic research, publication, and commentary, due to an unprecedented deadly terrorist campaign that killed thousands over the course of more than 356 attacks between 2001 and 2014 (Bamidele 2017). The Jama'atu Ahlis Sunna Lidda'awati wal-Jihad received significant criticism from civilians and the international community after these attacks, and eventually lost their support.

These arguments and cases imply that it is no good for rebels to victimize civilians to achieve their goals. Therefore, the following hypothesis is posited.

Hypothesis 2: Insurgents who have targeted civilians during a civil war are less likely to start a civil war again.

To summarize, civilians are eventually targeted and killed by both the government and rebel groups during civil war. First, violence used by the government causes civilians to criticize and question the legitimacy of the government, which eventually leads individuals to strike back against the hated government (Wood 2003). Thus, the possibility of recurrence increases after the end of the civil war.

Second, when rebels use violence, civilians entrust themselves with the protection of the government, which is stronger than the insurgents. This limits and weakens the support base of the rebels which is limited by certain regions, ethnicities, and religions, due to the nature of the rebels (Kalyvas 2006). Therefore, even if the civil war ends or continues, it will not receive civilian support and eventually cannot continue the civil war against the government.

TABLE 2 below presents the relevant hypotheses that will be empirically analyzed in the following parts.

Methodology

A survival analysis is conducted to examine the duration of time before an event occurs. Events such as “death,” “failure,” and “termination” are frequently used in medical areas as well as in the social sciences, whereas militarized disputes and unemployment (Box-Steffensmeier and Jones 2004). Unlike the regression or logit method, survival analysis is capable of measuring the variables that affect the dependent variable as time goes by. In the current research, the dependent variable would be “duration of prolonged peace” in the post-civil war period. In this regard, my research will trace down the failure-time process and examine the time for a civil war to recur under the given conditions.

Rather than using a regression model, our current research uses a survival analysis for the following reasons, based on findings by Box-Steffensmeier and Jones (2004). First, the duration can be calculated and put into a standard linear regression models, such as OLS regression. However, duration data always has a positive value because it counts time-to-event and substantial asymmetry exists among duration times. Second, although the author could alleviate this distortion problem by using a natural log, right-censored data cannot be treated appropriately in a linear regression model. In this study, multiple civil wars that present uncertainties in their exact termination times have been examined. Survival analysis distinguishes these right-censored observations from uncensored data, among which termination is completely observed in the study. Survival analysis properly applies their differences in the likelihood function, whereas a linear regression model cannot properly treat the censored problem.

Unit of Analysis and Data

To test the hypotheses, the research used a dataset with a list of intrastate wars from the UCDP version 21.1 (Uppsala Conflict Data Program), which comprises all the 407 cases of intrastate wars from 1946 to 2021. To track down the recurrence of civil wars, the author distinguished the opponents of civil war and coded only disputes with the same opponent as a recurring civil war because each party in the dispute has different interests and issues. Data on one-sided violence (OSV) against civilians from the UCDP One-sided Violence Dataset, version 21.1 were used to set our independent variables. Using this data, containing the best and high estimates, the number of civilian casualties in each conflict was compared and analyzed.

The datasets used in this paper are primarily derived from the UCDP version 21.1 (Uppsala Conflict Data Program) and others from the one-sided violence (OSV) against civilians from the UCDP One-sided Violence Dataset version 21.1, Polity 5, SIPRI, Correlates of War (COW), as well as Fortna (2004), Walter (2004), and James Fearon’s (2003) research on civil wars. To improve the research design, we referred to data from other scholars, including DeRouen and Bercovitch (2008), Fortna (2004), Walter (2004), Fearon (2003), and Ohmura (2011), on civil war recurrence for use in combination with UCDP data.⁶ The UCDP intrastate war data is defined as “a minimum of 25 battle-related deaths per year and per dyad” (Eck and Hultman 2007, 235). This dataset provides annual estimates of the number of civilians killed by actors; “Side A” is always government and “Side B” is always the “non-governmental group.”⁷ The data only includes direct and deliberate killings, which is of particular interest to this study, thereby excluding other forms of violence against civilians, such as unorganized violent riots, indirect deaths following starvation, civilians killed in cross fire, and civilians killed in battled-related activities (Eck and Hultman 2007).

This research considers an event as recurrence only if conflict breaks out between the same pair of protagonists. For example, Syria experienced five civil wars with four different insurgent groups.⁸ This suggests that among five civil wars, only one episode–conflict with SDF–is coded as civil war recurrence. When a civil war does not recur, the duration of peace is right-censored on the year of 2021. Based on this estimation of the dependent variable (peace duration), the 407 post-civil wars are at risk of recurrence (failure) for a total of 8,616 years. Among the observation of 407 cases of civil war, 123 civil war cases failed to maintain peace and fell back into a trap of conflict.

Dependent Variable

The dependent variable measures “the duration of time before an event occurs” (Box-Steffensmeier and Jones 2004). To specify again, the time duration to analyze is the period of peace year after the termination of previous civil war in all countries, and the event dealt with here is the recurrence of the civil war. In the place of death or failure as used in a wide range of survival analyses, the event used in this study is “recurrence of civil war.” Taken together, survival analysis even allows for the inclusion of cases with unknown termination time due to non-recurrence within the observation time—denoted as right censored data, the research codes 123 cases as “failure” of the event among 8,616 analyses of time. The mean duration of post-civil war is 21.20 years, and the median duration is 16 years.

Independent Variables

To examine the determinants of civil war recurrence based on the duration of post-civil war peace, this study suggests two independent variables: Government Targeting Civilians and Insurgent Targeting Civilians. The main two independent variables in this research are based on the UCDP One-sided Violence Dataset, version 21.1. One-sided violence data includes the use of armed force by the government of a state or by a formally organized group against civilians, which result in at least 25 deaths. In this study, high fatality estimation is used when the government or organized group intentionally victimized civilians in the previous civil war.⁹

Scholars on civil wars have used the estimation of 100 casualties to account for intentional killing during a period of sustained violence (Krmaric 2017). Based on that, the independent variables in this study are measured and combined with the dataset derived from Krmaric (2017) and Ulfelder and Valentino (2008) works on civilian targeting.

First, Government Targeting Civilians is a dummy variable indicating whether a government intentionally victimized civilians in the previous civil war. However, states may deal with multiple insurgent groups, and whether a government victimized civilians against a particular insurgent group was analyzed through identifying the location indicating where the episode of violence occurred. All the episodes are analyzed based on civilian causalities in each year of the previous civil war period. In this regard, the author coded “1” only when at least 100 noncombatant fatalities occurred in the previous civil war, and “0” otherwise.

Second, Insurgent Targeting Civilians is a dummy variable indicating whether Insurgents victimized civilians during the previous civil war. Insurgent groups operate in various regions; however, the study only took account of an event if it occurred in the same location. In this regard, the author coded “1” only when at least 100 intentional noncombatant fatalities occurred due to insurgent groups during the previous civil war, and “0” otherwise.

Control Variables

This study includes control variables which are considered as causes of civil war onset, and/ but may also become the causes of civil war recurrence. Therefore, joint democracy, military expenditure, total population, mountainous area, oil export, previous civil war duration and ethnic fractionalization, are included in the analysis models.

First, joint democracy is measured using the Polity 5 project, indicating a democracy-autocracy scale (-10 to 10) after a peaceful era begins (Marshall, Jaggers and Gurr 2015). The authors consider the country democratic if states score 6 or above using the Polity 2 variable; and coded it as “1,” and those scoring under 6 are coded “0.”¹⁰ Democratic states are less likely to experience a civil war again because it provides people with an opportunity to express their grievances (Fearon 2003; Mason et al. 2007; Ohmura 2011). Therefore, a high level of democracy is expected to lead to longer duration of peace following the end of civil war.

Second, the duration of the previous civil war impacts the risk of civil war recurrence. The duration of war varies from 1 day to 18,992 days.¹¹ This variable is also obtained from Fortna (2004) and Fearon (2003) and measured using the natural log of daily duration. For the most part, there are arguments presented by scholars that with limited resources in manpower and weaponry, the leaders and citizens of a nation are likely to avoid another civil war, by recognizing that war requires high levels of resources (Hartzell, Hoddie, and Rothchild 2001).

Third, military expenditure is measured using expenditure per capita from the SIPRI Military Expenditure Dataset. The effects of military forces may rely on their capability to deter or defeat renewed insurgency. Increasing military expenditure is likely to deter insurgency and civil war recurrence; however, it may also encourage one to become involved in more conflict (Fearon and Laitin 2003).

Fourth, log total population is measured using a natural log of population from the World Bank’s World Development Indicators. In particular, a large population may increase the risk of recurrence by facilitating insurgent mobilization, fueling grievances or complicating post-war peace (Walter 2004).

Fifth, mountainous area refers to the percentage of the country covered in mountains. Mountain data is derived from Fearon and Laitin (2003), who measured using percentage, and was also provided in Collier, Hoeffler and Söderbom (2004). The presence of Mountainous areas is positively related to the onset of civil war (Fearon and Laitin 2003) and is likely to enhance the capacity for survival of the insurgency: the ability of the rebels to exploit primary products for their benefit (i.e., viability), operate guerrilla warfare and even hide and set military bases in the mountains (Le Billon 2001).

Sixth, ethnic fractionalization can deepen hostility and facilitate insurgent mobilization. This study used Fearon and Laitin’s (2003) ethnic fractionalization index to examine the risk possibility of recurrence.

Seventh, oil export is an also important indicator for the risk of armed conflict related to natural resources and the variable is derived from Fearon and Laitin’s study (2003). Oil export is coded “1” when states receive more than one third of their export revenues from oil exports, otherwise “0” (see Appendix 2 for a detailed description of variables).

Kaplan-Meier Survival Curves

The Kaplan-Meier survival curve is used to compare and express the results of survival analysis for multiple groups without controlling for other covariates. Overlapping lines in the Kaplan-Meier survival curve graph demonstrate no significant difference across groups. The method is used to simply present and visually illustrate the different durability patterns in targeting civilians (FIGURES 1.1, 1.2).

FIGURE 1 illustrates the Kaplan-Meier survival graphs displaying the duration of peace after civil war depending on the use of violence against civilians during the previous civil war. FIGURES 1.1 and 1.2 were put together to compare the difference between the two actors’ behavior.

FIGURE 1.1 illustrates civil war recurrence more likely to occur when the government killed and victimized civilians during the previous civil war, whereas FIGURE 1.2 implies that insurgents killing and victimizing civilians during the previous civil war less likely to initiate civil war again. In addition, compared to FIGURE 1.1 showing large deviation in line, FIGURE 1.2 shows less deviation between the two lines, indicating that insurgencies which killed civilians during previous civil war, or did not, are both likely to start civil war again. However, FIGURE 1.2 illustrates the survival curve of insurgents killing and victimizing civilians in previous civil wars falls less rapidly than that of FIGURE 1.1 where governments killed civilians. This difference indicates that governments that killed and victimized civilians are more likely to fall back into a trap of conflict. Based on the survival graph presented, civilian killing demonstrates the dynamics working on both different sides in the recurrence of civil war. However, this could differ when other covariates are put together. Thus, the survival graphs will be shown again with the covariates included after presenting the cox hazard regression of the research.

To summarize, FIGURES 1 shows differences in the recurrence of civil war (failure event); negatively related when insurgents killed civilians, positively related when governments killed civilians during previous civil war.

FIGURE 1 shows the result of a simple analysis, indicating whether the event (civil war recurrence) occurs and if so, how long it takes for that event to occur depending on the resort to violence against civilians in the previous civil war. To examine the hypotheses more meticulously with other covariates controlled, the Cox method is used and displayed in TABLE 3.

Cox Proportional Hazards Models

The analysis uses a Cox proportional hazard model. Various models are used for survival analysis, but each different model assumes a different functional form for the hazard rate. In the current study, the distribution of civil war recurrence cannot be clearly expected; therefore, this study uses the Cox non-parametric proportional model (simply, the Cox model), which makes no assumption regarding distribution over time.

The main results from the Cox proportional hazard models for the two main independent variables are reported in TABLE 3 with control variables. All models ran a goodness-of-fit test to check whether the Cox model’s proportional hazard assumption fits the observations (see Appendix 3 for more detailed results).

The numbers in TABLE 3 are robust standard errors (in parentheses) clustered by conflict and hazard ratios for convenience of interpretation. A hazard ratio higher than 1 indicates that the increasing value of the covariate is associated with an increased risk of civil war recurrence, whereas a hazard ratio less than 1 indicates a decreased risk (Box-steffensmeier and Jones 2004).

The results provide evidence in support of all hypotheses claiming government killing civilians increases the risk of civil war recurrence, whereas insurgents killing civilians and mass killing civilians decrease the risk of civil war recurrence. Model 1 shows a substantially higher hazard of civil war recurrence associated with the governments targeting (killing) civilians variable. The results are statistically significant and robust to the inclusion of control variables in Model 1. The hazard ratio indicates that government killing civilians in previous civil war increase the risk of civil war recurrence by a 354.8% higher [=(4.548-1)*100] hazard rate and shorter peace time, compared with not killing civilians, holding covariates constant.

The effects of insurgents targeting (killing) civilians are also statistically significant in Models 2 with holding covariates. To be more specific, Model 2 shows a statistically significant decrease of 48.8% [=(0.512-1)*100] in the risk of civil war recurrence. When all two independent variables are included in Model 3, the coefficient on insurgents targeting (killing) civilians seems to lose statistical significance. However, the hazard ratio continuously holds under 1, thereby indicating it decreases the risk of recurrence, but is inconsistent in its statistical significance. In contrast, government targeting civilians variables hold statistical significance and even show stronger increase for government targeting civilians and decrease for mass killing in the hazard ratio of recurrence.

Building upon all the Models displayed in TABLE 3, H1 is supported by Model 1 demonstrating that government killing civilians are associated with increasing the risk of recurrence has strong statistical significance, which can be explained as government targeting civilians in previous civil war will likely fall back into the trap of conflict after the end of civil war. On the other hand, H2 is supported by Models 2 associated with decreasing the hazard ratio of recurrence. However, insurgents killing civilians seem to lose its statistical significance. This may be interpreted as mass killing and government targeting (killing) civilians holds stronger significance in the risk of recurrence.

Other control variables have a consistent direction of hazard ratio throughout all models. Moreover, mountainous, and military expenditure variables are statistically significant throughout all models. Military expenditure and mountainous variables are associated with increasing the risk of civil war recurrence. To be more specific, military expenditure may encourage people to engage in more battles with more military assets and eventually contribute to victimizing civilians during conflict. Mountainous areas may contribute to insurgency activities because mountainous areas are considered to be suitable for guerrilla warfare, so the insurgents likely continue to resist for a long time and cause civilian casualties. These two variables are also associated with genocide, which requires costly weapons to effectively carry out this strategy and uses them to suppress rebels operating in mountainous areas, which eventually leads to civilian casualties. Another interesting variable in relation to civilian casualties, previous civil war duration, shows statistical significance. Studies have shown that the long duration of civil wars is less likely to recur because both actors, including civilians, recognize that war requires high levels of resources. However, this study shows that a longer duration of the previous civil war increases the risk of recurrence. These results can be interpreted as increasing the risk of recurrence of civil war because the longer the civil war period, the higher the civilian casualties.

With all covariates included FIGURE 3 shows the likelihood of peace in a given peace year for each event of conflict with government targeting civilians, insurgents targeting civilians, respectively, compared to those without controlling other variables. FIGURES 3.1 and 3.2 show a minor difference in the first year, but government targeting civilians in previous civil war is roughly 20% likely to fall back into a trap of conflict in the second peace year, and nearly 30% likely in the seventh peace year. By contrast, the government that does not target civilians is roughly 10% likely to fall back into a trap of conflict in the second peace year, and nearly 20% likely in the seventh peace year. Thus, H1 is supported by FIGURE 3.1 controlling for other covariates.

On the other hand, in FIGURE 3.2, insurgents targeting civilians in the previous civil war is roughly 15% likely to result in the recurrence of civil war in the second peace year, and nearly 25% likely in the seventh peace year. Insurgents that do not target civilians are roughly 25% likely to result in the recurrence of civil war in the second peace year, and nearly 35% likely in the seventh peace year. FIGURE 3.2 in the survival function enforces H2 and includes covariates in the model.

Throughout this analysis, it has been shown that the governments killing civilians are most like to fall into the trap of conflict. In addition, insurgents that caused civilian casualties are not only undermining the legitimacy of the civil war, but also withdrawing support for the insurgents and participating in the opposition, thus reducing the possibility of a recurrence of civil war.

Case Selection

TABLE 4 presents a summary of the independent variables according to the case that best describes them. There are countless cases of civil war, but among them, cases were selected for the following two reasons. First, only cases existing in the data from 1945 to 2021 analyzed by this study were selected. Second, only the clarity of the data—that is, the estimate of civilian casualties—was considered.

The first example explains H1 that if the government uses violence against civilians, resistance will increase, and a recurrence of civil war will occur. The second case was selected as an example explaining H2, but instead, the opposite case was selected. As previously noted, if the key factor for the rebels in the civil war is the civilian support base, the case of strengthening the support base could have another outcome. If the existing arguments and the research claims are correct, the rebels who have strengthened their support base would have continued the civil war and received their demands from the government. For the third case, two cases were selected, and first, two different cases were selected to analyze whether genocide was effective in different environments and conditions. Then, let's look at the examples one by one in the next section.

Government Targeting Civilians: Case in Angola

In the year of 1975, three Angolan liberation movements succeeded in negotiating Angola’s independence from Portugal’s former colonial rulers (the Swedish Institute of International Affairs 2016a). The three movements were supposed to share power until they could prepare for elections, but hostility began to emerge between them, and a civil war broke out in 1975. The People’s Movement for Liberation of Angola (MPLA), one of the groups, took control of the capital and formed the government. The other two groups, the National Union for the Total Independence of Angola (UNITA) and Front for the Liberation of the Enclave of Cabinda (FLEC-FAC), did not accept the result and tried to take over the government. UNITA received support from the United States to resist the government, whereas FLEC-FAC quickly gained control of the countryside. In this situation, the MPLA, which took control of the city and established a government, tried to suppress insurgents with support from the Soviet Union. They drove each other into a long and repeated trap of conflict. In 1994, MPLA and UNITA signed a peace agreement with each other, but in 1998, civil war broke out again. During that period, the MPLA and FLEC-FAC fought a total of seven civil wars, with the government estimated to have killed at least 50,000 civilians during all conflicts (Valentino, Huth and Balch-Lindsay 2004).

The case showed the government taking the risk of killing civilians to suppress the insurgents, which led to stronger resistance. Many civilians were killed by the government, thus the number of civilians joining the rebels increased, and they went through stronger resistance and repeated civil war.

Insurgents’ Effort to Minimize Civilian Casualties: the Case of Indonesia

In this case, the study introduces a case wherein the rebels won against the government by putting effort to minimize the sacrifice of civilians. This case can be considered a successful case of insurgency wherein insurgents recaptured the government by winning the elections held after the peace treaty.

The civil war between Indonesia and the Free Aceh Movement (GAM) was a case wherein the Indonesian government used summary executions, arbitrary arrest, and the scorched-earth tactic to exterminate the GAM rebels. Civilians gave more support to the GAM against such a brutal government. According to UCDP data, only 243 civilian casualties reported on the GAM side alone during the civil war, and more than 2,000 civilian casualties were reported on the government side. The civil war, which began in 1973, lasted for 30 years and ended in 2005, and did not recur until the present day. In addition, it ended with an insurgent victory in the elections conducted after signing the peace treaty.

This example is very ideal and exemplary for insurgency, wherein the insurgents, whose support for civilians is important, eventually won the election and legitimately recaptured the government by drawing their support against the government. In this case, the possibility of a recurrence of civil war is reduced by a government that has secured legitimacy supported by the public. Therefore, the support of civilians is essential for the insurgency to win the civil war or to continue to wage civil war.

War between nations does not occur frequently. However, civil war has replaced it and is occurring frequently in many countries. In many civil wars, civilians are sacrificed and some thus become refugees. Civilians and refugees who are exposed to such violence either entrust themselves under the protection of the government for their own survival or join the rebels in opposition to the government. This study focused on the impact of these civilians on the recurrence of civil war. Many civil war studies have explained why they continue to occur. However, in these studies, the causes of recurrence were analyzed by focusing on the actors of the dispute or other factors. This current study differs from the literature in that civilians, which account for the largest proportion in civil war countries, were included.

The results show that violence against civilians influences the recurrence of civil war for both the government and rebels. The government should avoid violence against citizens to prevent further conflicts. The rebels, on the other hand, must win the support of the civilians to resist the government and strengthen their legitimacy. Therefore, it appears that violence against civilians should be reduced for the interest of both actors.

Violence in civil war gives birth to further violence and ensures that these states continuously fall into the trap of violence. This in turn results in the loss of numerous human and material resources. Therefore, it is necessary to prepare an institutional or international legal system that can prevent genocides in the international community. Today, armed conflicts often arise from a mixture of interstate and intrastate tensions, affecting regional stability and security above all else. Therefore, efforts by policymakers and scholars are required to reduce violence among civilians and possibilities of civil war recurrence.

However, further questions have remained unaddressed. First, in many civil wars, rebels unite to resist the government. This may be a temporary or permanent union, but it may also be separated at some point with different interests and issues. Due to this fluid change, there is a limit to tracking all insurgents and separating them according to their characteristics. Second, countries experiencing civil war have generally unstable administrative powers, and new governments often appear due to coups. In this case, the rebels and the government are intertwined with new issues and interests. Finally, the government tries to conceal their operations and data recognizing that killing civilians is barbaric and can be criticized. The rebels also do not track down all the violence they commit or clearly estimate casualties. In conclusion, it is difficult to track down the causes of all disputes due to the nature and complexity of the civil war. Therefore, continued efforts must be made to analyze each case to avoid falling into a trap of conflict.