WHEN LARGE CONFLICTS SUBSIDE: THE EBBS AND FLOWS OF VIOLENCE IN POST-SUHARTO INDONESIA

Abstract

The last decade has witnessed an extraordinary spate of scholarship on the ethno-communal violence that swept through Indonesia following the collapse of the Suharto regime. Yet we know very little about how these large-scale violent conflicts subsided and the patterns of post-conflict violence that have emerged since. We introduce evidence from an original dataset to show that the high violence period lasted till 2003, after which violence declined in intensity and scale. Despite this aggregate decline, we find that old conflict sites still exhibit relatively high levels of small-scale violence. We conclude that Indonesia has moved to a new, post-conflict phase where large-scale violence is infrequent, yet small-scale violence remains unabated, often taking on new forms. Finally, we propose that effective internal security interventions by the state are a key reason, although not the only reason, why large-scale violence has not emerged again despite the continued prevalence of low-level violence.

INTRODUCTION

The last decade has witnessed an extraordinary spate of literature on large-scale violence in Indonesia. For the most part, scholars have concentrated on explaining why violent conflicts emerged following the collapse of the New Order regime. As a consequence, we know a great deal about the gruesome civil wars and ethno-communal violence that swept the country between 1998 and 2003. Since then, there has been a remarkable decline in intensity and scale of violence. Yet we know very little about how and why these large-scale violent conflicts subsided. What are the patterns of de-escalation across conflict-affected areas? Do previous conflict sites continue to harbor old vulnerabilities, or are they subject to new forms of violence? How has Indonesia managed to avoid the re-emergence of large-scale violence during the past decade?

We answer these questions by examining evidence from the National Violence Monitoring System (NVMS), a new dataset that we created to provide the most comprehensive record of violence in Indonesia over a 15-year period. In doing so, we go beyond the article published in this journal some years ago, covering UNSFIR-2, the first large dataset on violence in Indonesia (Varshney, Tadjoeddin, and Panggabean 2008). Like Varshney et al. (2008), we primarily identify patterns of violence over time and across space since 1998. Unlike them, however, we designed our dataset to cover Indonesia's post-conflict period up to 2012, and we include sites of insurgency that were excluded from UNSFIR-2. Furthermore, our dataset adopts a broader definition of violence, while allowing for disaggregation of violence along multiple dimensions, including state responses. This enables us to analyze evolution in forms of violence over time.

We make three arguments. First, despite the dramatic reduction in overall levels of violence, areas previously affected by high levels of violence continue to harbor specific vulnerabilities. Old conflict sites exhibit significantly higher levels of collective violence compared to the rest of the country. However, while the frequency of violence in these areas has increased over time, the episodes of violence are now less deadly. Second, although a large proportion of violence in conflict-affected areas still takes the form of ethno-communal conflict, other interlinked forms of violence have also emerged. We conclude that areas previously caught in highly destructive violence have moved to a new phase where large-scale violence is absent yet small-scale violence remains unabated, often taking new forms. We are not confident whether an enduring equilibrium has set in. However, a puzzle has clearly emerged: given the continued vulnerability of post-conflict sites to small-scale violence, what has prevented the re-escalation of these sporadic events into larger conflagrations of violence?

The answer to this question leads us to our third argument. Though the primary purpose of this article is pattern identification, we do present a causal suggestion, which should be further explored. At this point in the evolution of our research, firm causal explanations, fully resolving problems of possible endogeneity, cannot be given. Causal suggestions, however, can be made. That is the form in which we are presenting our third argument.

The argument is that an improved response by the Indonesian security forces is a key, although not the sole, reason for the observed absence of reescalation of violence. Tajima (2014) has claimed that in the immediate post-Suharto period, security sector reforms such as the separation of the police from the military contributed to the inability and often unwillingness of state actors to intervene effectively. As a result, violence quickly escalated from small episodes into large-scale conflicts. We find considerable merit in Tajima's explanation, but we would like to propose an amendment, when we extend it to the post-conflict period (2004–2012). In this period, the security interventions were embedded in a new political settlement without which their effectiveness would have been doubtful. Only when the state began seriously to address the problem of large-scale violence in Indonesia did the civil wars and communal conflicts come to an end. Most significantly, the gradual consolidation of power by political actors and changing political will from the center allowed for a series of peace agreements. After 2004, this changed environment contributed to improvements in the incentives and ability of the security forces to prevent incidences of large-scale violence.

While we acknowledge the improved policing capacities of the Indonesian state, we also note that the police remain incapable of, or disinterested in, preventing smaller acts of violence. Thus, larger episodes of violence have been contained, though they have not been fully eradicated.

In short, although our post-2004 findings provide considerable support to Tajima's argument, we hesitate to present security interventions as a single-factor causal explanation. Security interventions, in any case, were embedded in a new political framework, which ought to be noted. It is quite possible that several factors caused the decline in violence. We are, in all probability, in a multivariate analytic universe. A fuller explanatory probe would be necessary, as the NVMS data and our pattern identification are analyzed further.

The paper is organized into four sections. First, we briefly review the literature on transition violence in Indonesia to identify critical gaps in our understanding of violence dynamics since 2004. Second, we introduce the National Violence Monitoring System (NVMS) dataset, which we created to overcome these empirical challenges. Third, we use the NVMS data to describe patterns of violence de-escalation and post-conflict violence in Indonesia. Drawing on both quantitative and qualitative data, we propose improved security interventions as a tentative explanation for the observed patterns. Finally, we conclude by noting the implications of our findings for a future agenda on violence research in Indonesia and beyond.

UNDERSTANDING POST-TRANSITION VIOLENCE IN INDONESIA: EMPIRICAL GAPS

Violence was a defining feature of the New Order regime for three decades. General Suharto came to power on the back of a massive anti-communist massacre (1965–1966), when over half a million people lost their lives (Cribb 1990; Robinson 1995; Heryanto 2006). Throughout the New Order years (1965–1998), the security arms of the state used violence to retain control, build power, and scare off challenges (Anderson 2001; Colombijn and Lindblad 2002). State-sponsored violence included the petrus (or “mysterious”) killings of alleged criminals from 1983–1985, when as many as 2,000 might have died (Bourchier 1990; Barker 1998; Siegel 1998). The military, which received a small share of its budget from official state sources, also used violence to generate income (Liem 2002). At times state repression and coercion led to armed resistance. Separatist violence ebbed and flowed in Papua from 1964 and in East Timor and Aceh from 1976 onwards, resulting in harsh counter-insurgency operations that killed many. Communal rioting also occurred, in particular in the late Suharto period, with outbreaks in Java (Sidel 2006), Kalimantan (Davidson 2009) and Sulawesi (McRae 2013).

But the violence that accompanied the fall of the New Order was new in both scale and nature (Klinken 2007). Two forms of violence—separatist civil war and communal violence—were especially deadly in the early post-transition violence. Long standing separatist conflict intensified in Aceh and East Timor (Aspinall 2009; Robinson 2001). Papua, too, saw separatist violence, although this was much more sporadic (ICG 2001; King 2004). Other provinces experienced massive communal violence. In contrast to the occasional outbursts of ethnic violence observed during the New Order, communal violence in the transition period continued for months or years, resulting in enormous destruction of lives and property. In Central Sulawesi and Maluku, the cleavage was primarily a religious one (McRae 2013; Klinken 2007); in North Maluku, ethnic violence morphed into inter-religious battles (Wilson 2008). Violence endured in North Maluku for almost a year; in Maluku and Central Sulawesi for multiple years. Extended communal violence also broke out in Indonesian Borneo (Klinken 2001; Davidson 2009). For around three weeks from late 1996 to early 1997, while Suharto was still in power, ethnic Dayaks in West Kalimantan attacked the migrant Madurese community; a second round of violence around two years later set Malays against the Madurese.

Beyond these areas of extended violent conflict, incidents of episodic large-scale violence also became more common. The biggest of these were the massive riots that engulfed Jakarta in mid-May 1998 and that preceded the fall of Suharto. Anti-Chinese riots also hit other Indonesian cities such as Medan in North Sumatra and Solo in Central Java (Purdey 2006; Panggabean and Smith 2011). In the 17 provinces for which data is available, an estimated 21,495 people lost their lives during the turbulent transition years of 1998 and 2003. ¹

Globally, a host of scholars have shown that violence often emerges during authoritarian breakdowns (Mansfield and Snyder 1995; Snyder 2000) and when rapid economic declines take place (Chua 2004; Paris 2004; Cramer 2006). In Indonesia, a rich set of theoretical and empirical approaches have been employed to explain the unprecedented outbreaks of large-scale violence during the early transition years. The period from 1998 to 2003 was indeed one of major national level political and socio-economic change driven by parallel transitions from authoritarianism to democracy (reformasi) and from a centralized to decentralized polity (desentralisasi) as well as changes to the structure of Indonesia's economy following the Asian financial crisis (krismon). These national-level changes constituted a “critical juncture,” as the rules of the game—what institutions would emerge, and whose interests they would represent—were redefined (Bertrand 2004). In this context, qualitative studies have pointed to the role of local leaders, who responded to these institutional shifts by using violence to cement or extend their control of the local political and economic arena (Klinken 2007). Others have examined the role of local anxieties about Islam's place in the new institutional set up that prompted religious leaders to assert their authority in a rapidly changing context (Sidel 2006).

Scholarship on early transition violence has also benefited from new quantitative datasets. The most comprehensive information about violence in Indonesia's early transition period came from the United Nations Support Facility for Indonesian Recovery (UNSFIR-2) dataset, which covers the 1990–2003 period (Varshney et al. 2008). Varshney examined newspaper reports of violence based on the methodology first used to build a dataset of riots in India (Varshney 2002). The UNSFIR dataset has generated a wealth of scholarship on the causal mechanisms that explain why violence erupted in some places and not others (Tadjoeddin 2004; Murshed and Tadjoeddin 2009; Østby et al. 2011; Tadjoeddin 2014).

Another dataset leveraged by scholars to explain post-transition violence is the Village Potential Survey (PODES) conducted by the Indonesian bureau of statistics since 2003. This nationwide survey asks village heads about violence that occurred in the past year and the impact it had. Micro-level variation in violence observed in the PODES has been used to advance arguments about the causal effect of weakened coercive capacity of the state—a result in part of the separation of the police from the military—on violence escalation (Tajima 2014). Other studies have used the same spatial variation in violence to confirm the importance of economic factors in post-transition violence (Barron, Kaiser, and Pradhan 2009).

While the bulk of quantitatively oriented scholarship has been devoted to explaining large, protracted events, several studies have noted the increase in small-scale or “routine” violence during Indonesia's early transition years. A sharp increase in the frequency of vigilante violence and group brawls in Java around 1998 has been seen as resulting from income inequalities exacerbated by the financial crisis (Tadjoeddin, Chowdhury and Murshed 2012). Others have demonstrated the link between high levels of mob lynching in typically peaceful provinces and intensification of local political competition during the initial implementation of decentralization reforms in 2001 (Welsh 2008).

Although the scale and protracted nature of early transition violence in Indonesia had led many observers to conclude that the risk of Balkans-style fragmentation was real (Emmerson 2000; Aspinall and Berger 2001), by 2004 a noticeable decline in violence set in. Relatively little is known about why these debilitating conflicts subsided and how patterns of violence have evolved since. Comparative experience suggests that areas that are exposed to high levels of violent conflict continue to be vulnerable to violence for long periods of time (Walter 2004). Even areas that do not experience a full-scale recurrence of old violence are likely to experience other forms of post-conflict violence (Rodgers 2009; Chaudhary and Suhrke 2008; Fortna 2008; Boyle 2014). In some cases, such as Nicaragua and Guatemala, the end of a major conflict is often accompanied by a rise in criminal forms of violence (Moser and McIlwaine 2001).

Extant studies provide tentative indications that many of the same dynamics are at play in post-conflict Indonesia as well. For instance, many of the causal factors associated with the outbreak of large-scale violence, such as political uncertainty and institutional weakness, began to subside after the initial turbulence that accompanied the fall of Suharto. Rainbow cabinets were formed, containing most elements of political society; the military, realizing that core interests were not under significant threat, also pledged loyalty to civilian political institutions (Crouch 2010; Mietzner 2009; Horowitz 2013). New rules of the game were forged that reduced the incentives of elites to support—actively or passively—the escalation of violence (Barron 2014). The gradual nature of these changes suggests that their impact on conflict de-escalation is likely to be temporally and spatially uneven rather than uniform. Furthermore, evidence from Maluku, Central Sulawesi and Aceh indicates that sporadic violence has continued to erupt in these regions even after a series of peace agreements marked the formal conclusion of hostilities (McRae 2013; Kloos 2014; ICG 2012a; Barron, Azca and Susdinarjanti, 2012). Studies also suggest that areas like Lampung and North Kalimantan, which had previously managed to avoid large-scale violence, have become increasingly more susceptible to ethnic violence (ICG 2010; Wilson 2013; Jones 2013; Barron and Madden 2004).

Despite these tentative indications, lack of systematic data on violence since 2003 has prevented an investigation of some basic questions about the de-escalation process and post-conflict dynamics. Did violence levels drop precipitously, or has the decline been more gradual? Has the process unfolded differently across different conflict sites? Has the decline in large-scale violence also been accompanied by a decrease in small-scale “routine” violence? How much violence is still taking place in these “post-conflict” areas? Is ethnic violence still the most predominant form of violence or have new issues surfaced in its wake? Are the old conflict sites still the most vulnerable, or has violence shifted to other areas? Our new dataset seeks to address these gaps.

INTRODUCING THE NATIONAL VIOLENCE MONITORING SYSTEM (NVMS) DATASET

The National Violence Monitoring System (NVMS) dataset records all incidents of violence in 16 provinces, which represent all major island groups and account for about 53 percent of Indonesia's population, as reported by over 120 local news sources. The selected provinces include the “high-conflict” provinces that were affected by large-scale violence following Indonesia's democratic transition as well as “low-conflict” areas that were not. For the high-conflict provinces—Aceh, Maluku, North Maluku, Central Sulawesi, Central Kalimantan, West Kalimantan, Papua and West Papua—data has been compiled since 1998. For East Java, North Sumatra, Lampung, East Nusa Tenggara, West Nusa Tenggara, South Sulawesi, North Sulawesi and the Greater Jakarta Area (Jabodetabek), classified as relatively low-conflict areas, data has been collected mostly since 2005. By 2012, the NVMS had recorded 30 distinct variables for 163,466 incidents, which collectively resulted in 36,222 deaths, 132,110 injuries, 75, 937 damaged buildings, 4,322 kidnappings and 22,529 sexual assaults. These indicators can be disaggregated at the provincial, district, and sub-district level. As far as we know, the NVMS is the largest dataset of violence created for any single country ² (a summary of the scope of NVMS dataset is provided in Appendix 1).

Constructing the NVMS

Putting the dataset together took four years and involved several steps, which ought to be briefly summarized. ³ The first involved deciding the scope of violence to be covered and the areas to include. Our final sample ensured that data included areas with high, medium, and lower levels of violence. While the data are not formally representative of all of Indonesia, the coverage of sites with high, medium, and low violence gives us confidence that the observed patterns are in all probability generalizable to the whole country.

The second step was to decide which sources to use in each province. Following UNSFIR's attempt, newspapers have been widely acknowledged as a reliable, if not perfect, source to collect violence data in Indonesia (Varshney 2008). Others have since devised ways to improve upon UNSFIR's work. For example, studies have shown that provincial newspapers, while better than the national ones, still significantly under-report levels of violence. Barron and Sharpe (2008) compared death tolls from UNSFIR-2 with those from a violence dataset using sub-provincial papers for 12 districts in two Indonesian provinces for 2001–2003. Employing the same definition as UNSFIR, they found three times as many deaths from collective violence. ⁴ These findings led us to conduct extensive source assessments in the selected provinces, aimed at mapping the availability and quality of both media and non-media sources. These assessments confirmed that subnational media, while not perfect, were the best source for information on violence in Indonesia.

To minimize the weaknesses of media sources, we adopted two strategies. First, multiple newspapers were collected from each province, based on a mapping of district coverage, violence reporting policies and political affiliations of media sources. This ensured that one could make up for the possible flaws of one newspaper by using another source. Newspapers with overt political biases and those that did not fact-check stories were entirely excluded. Second, where media coverage was low or reporting was dubious, non-media sources were used to supplant and triangulate the data; these included academic articles and books on violence in Indonesia and its provinces, policy papers including those from the International Crisis Group and Human Rights Watch, and events recorded in other datasets (including UNSFIR-2). Finally, we asked experts on particular conflicts to look at the data on their provinces to assess plausibility and to try to identify any inaccuracies.

Following source selection, researchers were sent to each province to collect all available archives by photographing each page of every newspaper. Over two million newspaper images were digitized during this process. Trained coding teams in Jakarta then combed the images and clipped articles related to incidents of violence. A standardized coding template was completed for each article. For each incident, over 30 variables were coded: when and where the incident took place; whether it was a crime or a conflict; its physical impact; ⁵ the actors involved; the issue that appeared to drive the violence; the form violence took; the weapons used, etc. Important for our purposes here, information was also coded on what interventions were taken to try to stop escalation and whether they were successful. Where articles reported different levels of violence and casualties, the more conservative figures were used. To ensure that the coding process was standardized, a detailed data manual was developed that specified systematic quality control procedures, including inter-coder reliability as well as reliability of the same coder over time. ⁶

What is New About the NVMS?

Extension of Temporal and Spatial Coverage

A major contribution of the NVMS dataset is that it extends the temporal coverage of available violence data in Indonesia to include the “post-conflict” period beyond 2004. As noted earlier, the UNSFIR database contains observations only until 2003, when most ethnic and religious violence was thought to have begun to subside. While other available datasets such as PODES have continued to track violence, the survey is only conducted every three years and thus contains large observation gaps. Because the NVMS records incidents of violence continuously between 1998 and 2012, the data allows us to track changes in patterns of violence over an extended period of time.

The broad spatial scope of NVMS also makes it more conducive to the analysis of post-conflict trends. We collect identical indicators of violence from both “high-conflict” provinces that were affected by large-scale violence and “low-conflict” areas that were not. This allows for a comparison of violence dynamics across the two samples, high and low, in order to assess the extent to which convergence has occurred in the post-conflict period. Finally, the NVMS is the first violence dataset that covers Aceh and Papua, sites of large-scale separatist violence. These provinces were previously excluded by other datasets, including UNSFIR-2, primarily because ongoing conflict in these areas posed security risks to research teams. Because the NVMS effort started after 2008 under a considerably improved security environment, our team was able to collect data from multiple local newspapers in both Papua and Aceh from 1998 until 2012.

Broad Definition of Violence and Disaggregation

Another contribution of the NVMS dataset is that it adopts a broad definition of violence and yet allows for disaggregation of violence along multiple dimensions. Most available datasets in Indonesia tend to focus on collecting information about large-scale violence. For instance, UNSFIR-2 limited its scope to the collection of incidents of “collective violence” while the PODES questionnaire similarly focuses on incidents of “massive clashes” (perkelahian masal). Other datasets have focused on specific issues such as electoral hostilities (Tadjoeddin 2014) or vigilante violence (Welsh 2008). While these sources have been invaluable in improving our understanding of violence in Indonesia over the years, they are limited in two ways.

First, a focus on collective events is likely to underestimate the full impact of violence in the context of armed conflict. Smaller scale incidents of inter-personal violence, such as crime and revenge killings, are usually excluded from the analysis of civil wars and communal violence because they are not explicitly connected with the political master-cleavage driving the larger conflict. However, this conceptual distinction between collective and private violence has been challenged by scholars who argue that armed conflicts transform violence into a “joint process [involving] the collective actors’ quest for power and the local actors’ quest for local advantage” (Kalyvas 2003, 486). Second, single-issue datasets do not permit investigation of how violence might evolve from one form into another—e.g., from a localized inter-personal fight over land ownership into a larger communal clash. ⁷ Existing scholarship points out that forms of violence can change over time (Sidel 2006; Davidson 2009).

To overcome these limitations, we use an inclusive definition of violence in the NVMS: we include all incidents of violence where intentional physical damage is done to persons or property. Incidents are classified into four main categories: conflict, crime, domestic violence, and violence used by law-enforcement agencies. ⁸ Incidents of conflict are further categorized by the type of dispute: for example, resource-related, governance-related, electoral, separatist, ethno-communal, or vigilante violence. ⁹ Incidents can also be classified based on the scale of mobilization. “Collective” violence is defined as those incidents that involve the mobilization of a group of ten or more individuals to participate in a riot or a clash. In contrast, incidents of “interpersonal” violence are those that involve a smaller number of individuals and which are more spontaneous in nature.

Using a broader definition of violence also required using more extensive source materials. In contrast to the 19 provincial papers used by the UNFIR-2 dataset to collect incidents of large-scale violence from 14 provinces, the NVMS dataset uses 123 provincial and sub-provincial newspapers to cover 16 provinces. Table 1 presents a comparison of results from the NVMS and UNSFIR-2 in the provinces that were most affected by conflict between 1998 and 2003. Based on the use of a broader definition and additional sources, the NVMS dataset records 20 percent more deaths (9,985) compared to UNSFIR-2 (7,639) from the provinces that experienced the highest levels of ethno-communal violence: Maluku, North Maluku, Central Sulawesi and Central and West Kalimantan. Injuries and buildings damaged/destroyed are also higher in NVMS. The inclusion of separatist violence from Aceh and Papua more than doubles the previously estimated death toll from the early transition violence.

Table 1 Comparison of UNSFIR and NVMS Estimates of Early Transition Violence in Indonesia (1998–2003)

Data from: UNSFIR-2 and NVMS

* Aceh figures cover 1998–2005

** West Kalimantan figures cover 1997–2003

† The totals do not include estimated 1,485–1,585 deaths from East Timor (CAVR 2005)

This comparison offers two important correctives to existing accounts of early transition violence in Indonesia. First, while UNSFIR-2 found that 89 percent of all fatalities during this period came from ethno-communal violence, the inclusion of Aceh and Papua reveals that that separatist violence has in fact been just as deadly, accounting for nearly half of all fatalities. Specifically, out of the 21,067 fatalities recorded in the NVMS dataset between 1998 and 2005, 41 percent are associated with ethno-communal violence, another 40 percent are associated with separatist conflicts, and the rest result from smaller-scale, inter-personal violence. Second, using a broad definition of violence, the NVMS data shows that other types of violence related to electoral disputes, resource issues, popular justice and crime account for nearly 20 percent of all fatalities even at the peak of communal conflict in Indonesia. This important finding lends empirical support to theoretical calls for incorporating localized forms of violence into the studies of armed conflict (Kalyvas 2003).

Improved Accuracy

A final goal of the NVMS data collection effort was to overcome the accuracy challenges associated with existing sources. The local media sources are not perfect, but they are, in our assessment, the best available source of information. Accordingly, as already reported, we took several steps to improve accuracy, such as using multiple newspapers from the same province and using non-media sources to fill in gaps. These procedures yielded clear information benefits. For instance, we found that due to its reliance on individual village officials’ memory and reporting incentives, PODES data appears to overestimate fatalities in areas with large-scale violence. The 2003 survey—which provides data on violence between September 2001 and August 2002—reported that of the 4,849 people who died from conflict across Indonesia, 4,106 lost their lives in the high violence provinces of West and Central Kalimantan, Central Sulawesi, Maluku, North Maluku and Aceh (Barron et al. 2009). NVMS data for the same provinces records 3,415 deaths from violence. In contrast, it appears that in lower-conflict areas PODES under-reports violence. ¹⁰ While the 2005 PODES reported that just 276 people were killed in violent conflict nationwide (Vothknecht and Sumarto 2011), NVMS data for the same period found 1,207 deaths from conflict for just 16 Indonesian provinces, half the Indonesian total. ¹¹ Other official sources used in global assessments such as police data also significantly under-report violence. A comparison of police violence statistics in the Greater Jakarta area found that the former underestimated murders by 80 percent and rapes by 65 percent. ¹² In areas where police capacity is lower than in Indonesia's capital, police data are likely to miss even more incidents. ¹³

PATTERNS OF DE-ESCALATION AND POST-CONFLICT VIOLENCE IN INDONESIA

De-escalation: Omnipresent but Different Across Areas

Figure 1 shows the decline in violent deaths in 15 provinces for which we have time series data until 2009. The big wave of violence peaked in 1999 when the toll stood at over 4,500 deaths. The following two years saw similar levels of violence but by early 2002 the death toll began to decline as peace accords were concluded in Central Sulawesi and Maluku. By 2003, annual fatalities had reduced by half. In our judgment, the end of 2003 is a distinctive cut-off point to distinguish the high-conflict phase from the low-conflict one. The clear exception to this is the civil war in Aceh, which was only concluded in August 2005. As we know, all cut-offs in empirical analysis are to some extent arbitrary. However, by the end of 2003, it seems clear that a new phase of lower levels of conflict had definitely set in. As such, we refer to the 2004–2012 period as the post-conflict phase. ¹⁴

Data from: NVMS and UNSFIR-2. Data are for Central Kalimantan, West Kalimantan, Maluku, North Maluku, Aceh, Central Sulawesi, East Nusa Tenggara, Papua, West Papua, Banten, Jakarta, West Java, East Java, Central Java, West Nusa Tenggara and South Sulawesi

Figure 1 Decline of Violence-Related Deaths in Indonesia

While the overall decline of violence has been dramatic across the country, the process of de-escalation has not been uniform. Figure 2 shows patterns of de-escalation in different parts of Indonesia. We observe a sharp drop in violence followed by uniformly low levels of violence throughout the post-conflict period in areas where the capacities of warring groups for violence were destroyed (North Maluku). Aceh experienced a similarly rapid de-escalation of civil war violence following the Helsinki Memorandum of Agreement (MoU) between Gerakan Aceh Merdeka (GAM) and the government. However, new types of violence, most notably related to local elections and resource distribution, have since appeared in Aceh (Barron 2014). The de-escalation process has been more gradual in other areas.

Data from: NVMS

Figure 2 Patterns of De-Escalation in Indonesia: Decline in Violence-Related Deaths

Central Sulawesi continued to experience outbreaks of religious violence for several years even after a government-brokered peace agreement was signed in December 2001 (McRae 2013). West Kalimantan witnessed two episodes of anti-Madurese violence in 1997 and then 1999 before it finally subsided. Even as the pace of de-escalation has varied between provinces, all high-conflict areas in Indonesia have one feature in common: despite some experiencing frequent outbursts of post-conflict violence, a full reescalation to the levels of the early post-Suharto period has not occurred. A new phase has come about, in which violence continues to occur but does not rise beyond a particular threshold.

Post-Conflict Puzzle: High Frequency, Low Fatalities

Levels of Post-Conflict Violence and its Impact

Compared to the conflict period (1998–2003), the post-conflict period (2004–2012) saw a 79 percent reduction in annual violent deaths in the provinces previously ravaged by communal conflict. Aceh has seen a 94 percent decline in annual fatalities since the civil war was brought to an end in August 2005. ¹⁵ With the exception of injuries in areas of previous communal conflict, other impacts from violence have also declined (Table 2).

Table 2 Decline in Impact of Violence in Indonesia

Data from: NVMS

Given the steep decline, is it reasonable to conclude that violence levels in these areas have normalized? By normalization, we mean returning to all-Indonesia averages. To examine the degree of convergence with the rest of the country, we compared homicide rates ¹⁶ in the high- and low-conflict provinces. The data indicate that in the post-conflict period, high-conflict provinces still experience a homicide rate 32 percent higher than that in the low-conflict ones.

If we narrow our focus to look only at collective violence, defined in the dataset as those incidents that involve the mobilization of a group of ten or more individuals to participate in a riot or a clash, we also see continued differences between high-conflict and low-conflict provinces. We find that the frequency of collective violence is significantly higher in the former (Figure 3). The fatality rate resulting from collective violence—the ratio of deaths per incident—is seven times higher in high-conflict provinces than low-conflict ones in the post-conflict period. ¹⁷ It is also worth noting that incidents of collective violence in post-conflict areas have been on the rise since 2006, though deaths remain low (Figure 4).

Data from: NVMS

Figure 3 Collective Violence in Post-Conflict Period (Annual Avg./100,000 People)

Data from: NVMS

Figure 4 Collective Violence Over Tim in High-Conflict Provinces

Sites of Post-Conflict Violence

Highly localized concentration of violence was one of the defining features of Indonesia's high-conflict period. The UNSFIR study concluded that between 1990 and 2003, 85 percent of collective violence deaths were concentrated in 15 districts that represented just 6.5 percent of Indonesia's population (Varshney, Tadjoeddin, and Panggabean 2008). To examine how concentration has changed in the post-conflict period, we restricted our sample of violent deaths to only those that result from incidents of collective violence, and we included data from all 16 high- and low-conflict provinces to form a larger sample. ¹⁸ Using the UNSFIR definition, we find that in the post-conflict period, districts that represent just over 13 percent of the country's population account for 85 percent of collective violence deaths. ¹⁹ Collective violence is still quite concentrated.

Within the sample of high-conflict provinces, we find that not only is violence still highly concentrated in these areas but that it also tends to occur in the same sites that experienced the highest levels of violence during the conflict period. Figure 5 compares the share of violence-related deaths that occurred in the four most violent districts in each province in the conflict period with the share of deaths in the same districts in the post-conflict period. With the exception of North Maluku, the most violent districts in the conflict period still account for more than 40 percent of the total deaths in the province. In some cases, such as Maluku, Papua and Aceh, there has been virtually no change in the share of violence accounted for by these areas. This is consistent with findings from other parts of the world that show that areas that have experienced large-scale violence are likely to remain vulnerable to new, often interlinked forms of violence in the post-conflict period. ²⁰

Data from: NVMS

Figure 5 Post-Conflict Violence Concentrated in Old Conflict Sites

Composition of Post-Conflict Violence

Is the new violence merely a continuation of the old conflicts at lower levels, or have new types of violence emerged? Figure 6 shows the composition of violence in the eight high-conflict provinces during the post-conflict period. ²¹ Compared with the early post-Suharto years which were dominated by ethno-communal and separatist violence, in the post-conflict period some of these types of violence have continued but new ones have also emerged.

Data from: NVMS

Figure 6 Composition of Post-Conflict Violence in High-Conflict Provinces

Ethno-communal violence still accounts for a quarter of all conflict deaths in formerly high-conflict areas in the post-conflict period. This suggests that the communal grievances that led to large-scale violence in the early transition years have persisted, but now lead to small-scale forms of violence. Continuation of low-level religious violence in Poso and Ambon, intensification of tribal warfare in Papua, and the rise of village rivalries explain why this is still the category of conflict with the greatest impacts. Even though normal life has long been restored in (most of) these areas, small incidents involving members of different religious groups, or even rumors of such an incident, can trigger violent reaction. ²²

The post-conflict period in Indonesia has been accompanied by rapid economic growth and political and institutional change including direct local elections and decentralization. One consequence has been a rise in resource-related violence, mostly land disputes, which have increased from an average of 76 incidents a year in the conflict period to 158 in the post conflict period. ²³ According to some observers, land disputes between groups and individuals have existed for decades (The Habibie Center 2012). However, the recent inflow of outside investment for palm oil plantations and mining projects has raised the stakes in these disputes. This has resulted in a spate of violent confrontations between communities, private company employees, and even government officials. ²⁴

Local electoral violence has also risen, particularly in areas with a history of separatism. In Aceh, for example, the number of violent electoral incidents increased from an average of 7 a year (in the conflict period) to 47 per year as a result of fierce political competition between rival factions of the former rebel group (ICG 2012a; Barron 2014). In fact, the NVMS data shows that following the decline of the separatist violence, other forms of violence, including criminal activity, have increased in Aceh tending to peak around local elections (Figure 7). Papua and West Papua provinces have also seen a steady rise of electoral violence against the backdrop of an ongoing insurgency. Between 2008 and 2012, 48 deaths were recorded as a result of disputes during local elections, making Papua the site of the most violent elections in the country (NVMS).

Data from: NVMS

Figure 7 Violence in Aceh Over Time

Towards an Explanation: Improved Security Responses

What explains the emergence of this new phase in which large conflagrations of violence have been prevented, despite the persistence of these vulnerabilities in old conflict sites? There are a number of candidate explanations including the arrival of local elections, the decentralization of power, the creation of new provinces and districts, population movements and the creation and consolidation of a new national model with rules of the game that made large-scale escalated violent conflict unacceptable. We do not seek to rule out these explanations, and indeed elements of them may be linked to our causal proposal. Our data, however, shows that changed responses from security agencies to incidents of violence have played a significant role in determining current patterns of violence.

Tajima (2014) has argued that an absence of effective security interventions led to the escalation of small-scale violence into larger episodes during the early transition years. According to him, violence erupted in specific sites due to a mismatch in the coercive power of formal and informal institutions. During the turbulent transition years, when the state's security apparatus was disrupted, areas that were highly dependent on the state for maintaining communal order became most vulnerable to the violent expression of communal grievances. Tajima used evidence from multiple provinces to show how delayed responses by security forces to initial incidents of collective violence resulted in larger scale events.

Following the same line of reasoning for the post-conflict period, McRae (2013), Wilson (2013) and ICG (2011; 2012a) have suggested that the eruption of episodic violence in post-conflict areas is usually met with a stern response, often from the highest-level officials in the central government, translating into swift security interventions on the ground. In many cases, local police units have been quickly reinforced by the rapid deployment of regional reserves. In extreme cases, assistance is sought from the stand-by military units that have been stationed in these areas since the initial conflict.

We analyzed our data to check whether more effective deployment of security personnel has indeed prevented the escalation of violence. Beyond collecting information on incidents of violence, their impacts and triggers, the NVMS also records where an “intervention” was made to stop the violence during the course of an incident. If made, the identity of the intervener was coded (such as the law enforcement agency or civilian leaders) along with the result of the intervention. An intervention was coded as successful if the intervener was able to stop the violence and disperse the actors involved (through arrests or other means). Our data support these insights, indicating that active management of the security situation has contributed to the prevention of massive casualties—thus avoiding a full-scale reescalation to pre-2004 levels.

Figure 8 shows how both interventions by security forces and success have increased over time in the provinces that saw large-scale violence. Overall, incidents of smaller-scale sporadic violence (corresponding with the left axis indicated by shaded bars) have remained high over the years. However, the number of collective violence incidents (corresponding with the right axis indicated by solid black line) has risen steadily following an initial de-escalation around 2004. Incidents of collective violence tripled between 2006 and 2012. As we showed earlier, this surge in collective violence incidents has been accompanied by a sharp increase in the number of injuries and damaged buildings but not in the number of deaths, which has remained low.

Data from: NVMS

Figure 8 Change in Incidence of Collective Violence in High-Conflict Areas and Interventions by Security Forces

This paradox may be explained by the rise in interventions by security forces that have accompanied the upsurge in collective violence incidents. During the conflict period, only ten percent of these incidents saw any interventions from the security forces. Following 2006, however, security forces have intervened in about half of all collective violence incidents (corresponding with the right axis indicated by the dashed black line). Importantly, the rise in the incidence of collective violence has been accompanied by a proportional rise in security interventions. While the majority of security interventions were attempted by the Indonesian military during the conflict period, in the post-conflict period the police are the primary responder, accounting for over 80 percent of all interventions. The success of interventions has also risen steeply with an average success rate of over 85 percent (corresponding with the right axis indicated by solid grey line). In short, intervention by the security forces in collective violence incidents has increased and appears to have helped limit further violent escalation.

Why would the security forces intervene more often, and more effectively, in formerly high-conflict areas than in the past? Our quantitative data does not, and cannot, confirm the causal mechanism in place. However, other work, employing a process tracing methodology, has argued that such changes are most likely a result of the alteration in the national-level rules of the game, where a number of processes saw the emergence of widely shared institutional norms outlawing, or reducing incentives for, the use of escalated violence for political gain (Barron 2014). Peace agreements between contenting groups were especially important.

While we attribute the relatively lower levels of fatalities substantially to increasing and improved security interventions, we are not making any claims about the long-term effectiveness of this strategy. The number of interventions only represents attempts to halt the violence in incidents that have already occurred. It does not tell us anything about the attempts to prevent incidents from occurring in the first place. The Indonesian security agencies have been widely criticized for not following up security action with broader law enforcement measures such as the arrest of perpetrators or control of smaller-scale violence (ICG 2012b).

Further, while we claim that this damage-control approach is substantially responsible for keeping fatality levels relatively low in high-conflict areas, we are by no means implying that interventions by Indonesian security forces are adequate across the universe of cases. Of late, much has been written in the local and international press about the unwillingness of local police forces to intervene and stop violence against minorities. Documentary evidence has surfaced in many cases where mobs have attacked members of the Ahmadiyya community, while police officers have looked on. ²⁵ Indeed when we check the rate of interventions in low-conflict areas, we find that it is 30 percent lower than the rate in the high-conflict areas. The strategy of swift intervention in response to collective violence incidents in high-risk areas is not replicated in areas that have not seen large-scale conflict in the past.

Finally, we are not making any claims about the sustainability of this approach to violence management in the future. In fact, when we note that violence continues to be widespread in areas with an acute history of violence, the fragility of the current equilibrium becomes apparent. That violence is being managed through a series of successful security interventions is no guarantee that this strategy will work in the future.

CONCLUSION

By way of conclusion, let us recapitulate our principal arguments. Based on our analysis of the NVMS data, we have identified three key features of violence in Indonesia today. First, we show that the country has witnessed a significant decline in violence since 2004, although the process of de-escalation has been uneven across cases. Second, we find that while large-scale violence is precipitously lower than in the immediate post-Suharto years (1998–2003), small-scale violence continues in the post-conflict period. Incidents of violence have remained high, but fatalities per incident have come down. Specifically, provinces that experienced high levels of violence in the early period continue, in the later period, to be more violent than provinces that previously witnessed low levels of violence. We find no evidence of convergence to the national mean. Third, our findings reveal a paradox that marks the de-escalation process and the subsequent post-conflict period in Indonesia: despite the persistent vulnerability of post-conflict sites to small-scale violence, re-escalation of these sporadic events into larger conflagrations has been avoided. We argue that effective security interventions by the state, embedded in a new political settlement, especially in the high-conflict provinces, have played an important role in maintaining this equilibrium. However, it remains unclear whether this strategy will work in the long run unless the polity addresses the underlying causes of conflict. A deeper causal analysis is needed to fully understand the factors that drive observed patterns. Future work would do well to use the NVMS data to seek to confirm how our casual proposal holds up against other possible explanations.

The data and analysis presented here also have two implications for the study of political violence beyond Indonesia. First, the evidence from Indonesia suggests the need for further research on post-conflict violence. While there is a growing literature on how civil wars end and why they recur, such studies have tended to conceptualize periods of civil war and peace as dichotomous states. This prevents consideration of the forms of violence that often emerge in “post-conflict” situations. As a consequence, we know little about how spatial units marked by civil war or extended communal violence move towards relative peace and what prevents reescalation. The evidence we present from Indonesia shows that the “peaceful” phase, after large-scale violence ends, can also have a lot of violence. Second, the NVMS underscores the utility of using an inclusive definition of violence while maintaining the ability to disaggregate along multiple dimensions. Studies of political violence have tended to concentrate on one kind of violence (for example, Muslim–Christian, Sinhala–Tamil, Hindu–Muslim, Chinese–pribumi, etc.), or one form of violence (for example, riots or civil wars). Collecting data on all forms of violence together allows one to see how forms of violence can change over time, even when a period of peace, as we normally understand it, sets in.