Of the 408 communal conflicts that occurred around the world between 1989 and 2013, 304 erupted in anocracies, seventy in democracies and twenty-five in autocracies.Footnote 1 Yet, most of what we know about ethnic riotsFootnote 2 has relied on studies of clashes in mature democracies: Hindu-Muslim riots in India,Footnote 3 race riots in the United States,Footnote 4 and anti-migrant riots in France,Footnote 5 GermanyFootnote 6 and Britain.Footnote 7 Because mature democracies do not face the same level of uncertainty,Footnote 8 weakened institutionsFootnote 9 and heightened distributive demandsFootnote 10 as new democracies, an explanation that takes into consideration these differences is necessary.
This article contributes to the literature by examining ethnic riots in Indonesia from the final years of President Soeharto’s authoritarian regime through its first few years of democracy. Using district-level data from 1990 to 2005, I argue that ethnic riots tend to occur in ethnically divided districts with low electoral competition, because former regime party electoral dominance signals continued entrenchment and exclusion of opposition candidates at the local level, despite overall democratization at the national level. I argue that disgruntled local elites mobilize violence after disappointing elections in the first years of democracy in order to leverage their presence and alter the configuration of power. This argument implies that riots in democratizing countries are more likely to happen after uncompetitive elections that favor the outgoing regime’s party, and that riots would cease once these demands for inclusion have been met. Consistent with this idea, I find that every 1 per cent increase in electoral competition leads to a 1 per cent drop in the rate of riots. The dominance of Golongan Karya (Golkar) – the party associated with Soeharto – increases the likelihood of violence. Every 1 per cent increase in Golkar vote share leads to six times more clashes. Considering the rarity and costliness of riots, this effect is noteworthy. The less competitive a district and the higher its support for Golkar, the greater the level of violence it experiences, even after accounting for factors typically associated with rioting. Furthermore, I also find that the year after an election is more likely to be riotous than the year of or year prior to an election, and that districts in which turnover in favor of opposition candidates occurred are less likely to have clashes.
This article advances the literatures on ethnic riots, democratic transition and Indonesian politics. First, it assumes that the distinctive qualities of new democracies will affect local actors’ incentives differently, and derives directly orthogonal implications from what Wilkinson has found in Hindu-Muslim riots in India. Secondly, this article borrows the exclusion logic that Wimmer and others developed to explain armed rebellions globally, and applies it to ethnic riots.Footnote 11 To my knowledge, this is the first study that has tested this idea against Wilkinson’s minimum winning coalition idea to account for riots in democratizing countries.Footnote 12 Thirdly, it explicitly establishes a link between Golkar dominance and the onset of riots in districts in Indonesia, and expands the literature on Indonesian politics by focusing on election-related motivations for ethnic riots in democratic transition.
I would like to acknowledge some limitations at the outset. This article is limited by the lack of a direct measure of ethnic political configuration in the districts. While measures of the competitiveness of elections and Golkar dominance capture the challenges facing opposition candidates, they do not directly address the political representation of ethnic groups. A more rigorous test of the hypotheses outlined in this article would require much more fine-grained data that is currently unavailable. Secondly, this analysis treats ethnic and religious violence as a single analytical category. Although some have argued that ethnicity and religion are distinctive categories and that religiously motivated violence may work very differently from ethnic clashes,Footnote 13 in my argument ethnic groups are useful to local notables in so far as they provide alternative networks that can be mobilized, irrespective of which dimension of identity is being politicized. The ‘mobilizability’ of a group depends on a number of factors, one of the most important of which is the group’s size relative to the rest of the population.Footnote 14 Thirdly, this article does not test the micro-level dynamics linking elections and riots in new democracies. While my findings confirm the relationship between uncompetitiveness of elections and rioting, they do not address who riots, how elections function as a co-ordination point for mobilization, or whether there are qualitative differences between riots that happened before and after elections. With the growth of data collection work on Indonesia, I leave these questions for future study.
EXPLAINING ETHNIC RIOTS IN DEMOCRATIC TRANSITION
In explaining Hindu-Muslim riots in India, Wilkinson emphasizes the importance of electoral competition in local elites’ decisions to foment or quell riots.Footnote 15 Anticipating fierce competition, local elites allow riots to unfold in order to prevent target voters from supporting rival candidates. In states where elites rely on minority voters, however, violence is usually quickly quelled. In other words, ethnic riots are a campaign tool that activates voters’ ethnic loyalties and prevents them from supporting non-co-ethnic candidates.Footnote 16 They are used prospectively and strategically to mobilize support.Footnote 17
India has been a mature democracy for decades, however. Similarities between liberal democracies and other regimes notwithstanding,Footnote 18 new democracies have more uncertainties and greater distributive demands.Footnote 19 Given the temporarily weakened security capacity, for example, acts of violence that would be penalized in times of stability may go unpunished in the initial years of democracy.Footnote 20 Individuals in new democracies may commit violence as an alternative form of political engagement, in light of the country’s ineffective institutions.Footnote 21 Although distributive pressures exist in mature democracies, they are particularly intense in new democracies because outgoing elites try to secure their interests during the democratization processFootnote 22 while excluded individuals attempt to gain power. Consequently, initial election results may have symbolic significance and implications about future political configuration, and may affect violence differently in new democracies.
Wright finds that low levels of political competition at the onset of democracy predict political instability and regime collapse.Footnote 23 New democracies with low levels of political competition are more likely to experience civil conflicts, because those who were excluded try to subvert the regime. Implicit in this argument are: (1) the role of political exclusion as an underlying mechanism and (2) the use of low levels of political competition in new democracies as a signal of continued exclusion. One implication of this argument is that subnationally, clashes will erupt in areas where political exclusion is high. Since Wright’s analysis was done cross-nationally, his findings did not address this issue.
I argue that riots in transitioning Indonesia are driven by low electoral competition in ethnically divided areas. In the aftermath of a disappointing election, disgruntled elites turn to alternative networks – which in ethnically divided areas tend to be organized along ethnic lines – and generate violence to leverage their presence and pave the way for future inclusion. In transitioning countries where opposition parties may be weak, and where only groups affiliated with the authoritarian regime can have a meaningful presence at the local level,Footnote 24 ethnic groups are resources that local elites can easily mobilize. Uncompetitiveness would aggrieve excluded local elites anywhere, but only where groups are large enough would it be beneficial for local elites to reframe this grievance in ethnic terms. As such, even though exclusion occurs along party lines, the grievance over regime entrenchment assumes an ethnic manifestation and produces ethnic riots in divided districts.
My argument builds on the idea that the political exclusion of ethnic groups creates resentment and incentivizes groups to fight.Footnote 25 Examining all politically relevant ethnic groups globally from 1946 through 2005, Cederman, Min and Wimmer find that ethnic groups’ representation in national executive leadership positions predicts the onset of civil wars.Footnote 26 Excluded groups are more likely to challenge the ruling regime with violence than those that are represented in politics. Although they do not examine ethnic riots specifically, Cederman, Min and Wimmer’s basic idea that excluded groups will resort to violence in order to alter the existing power constellation is applicable to riots as well.
This idea bears some resemblance to Wilkinson’s logic on riots in India, but has directly orthogonal implications. Here, as in Wilkinson’s argument, riots are an alternative form of political engagement, which are utilized to place co-ethnics in political positions. The similarity ends there, however.
In Wilkinson’s formulation, riots are ‘brutal and effective form of campaign expenditure, designed by politicians to solidify ethnic majorities and diminish the importance of other politically relevant identities – especially in marginal constituencies and among pivotal groups of undecided voters – in the run up to elections’.Footnote 27 In the exclusion logic, riots are a reaction to disappointing election results that signal half-hearted reform and the continued exclusion of disgruntled elites. The argument has different implications for the type of elections that generate riots, the perpetrators, the timing of onset and the end of violence. In Wilkinson’s framework, highly competitive elections would incentivize elites to mobilize their co-ethnics. In the exclusion framework, uncompetitive elections are the precursors to violence.
Concerning the perpetrators, Wilkinson’s logic would imply that the rioters would be affiliated with the ruling elites, who can pressure security personnel to intervene or allow the riots to continue: the elites manipulate the riots because they want to stay in power. In the argument advanced in this article, the rioters would be the losers – the ones who wanted to enter politics via elections but failed.
This argument also entails a different implication about when violence ends. In Wilkinson’s story of competitiveness as a precursor to riots, violence ends when the incumbent wins. In my argument, violence ends when demands for inclusion are accommodated. I discuss these implications in greater detail in the subsequent section.
OBSERVABLE IMPLICATIONS
A few implications should follow. First, districts with low electoral competition are likely to experience more riots than competitive districts, because uncompetitive districts would present greater barriers for opposition candidates.
Hypothesis 1: Districts with low electoral competition will experience more riots than other districts.
It should also follow that districts in which the ruling regime continues to be dominant are likely to experience more violence. A large vote share of the regime party would indicate that the ruling party enjoys this advantage.
Hypothesis 2: Districts dominated by the party associated with the prior regime are more likely to experience riots than districts where the same party is less popular.
Because elections are the arena in which political competition occurs, proximity to election day should also influence a district’s likelihood of rioting. According to Wilkinson’s theory, the period leading up to an election should be particularly riotous. In my theory, because riots are an expression of disappointment over results that signal continued exclusion, riots should generally occur after elections, not before.
Hypothesis 3: Violence is more likely after an election, not before.
If riots are primarily driven by excluded politicians’ desire for elected positions, then once a turnover occurs, riots should decline, since demands for inclusion were met.
Hypothesis 4a: Districts with prior riots that have experienced an electoral turnover should see a decline in violence.
Hypothesis 4b: Districts with prior riots that have experienced increased electoral competitiveness over time should experience a decline in violence.
Another implication is that the competitiveness of elections should not matter during authoritarian years, since there would be no expectation that voting would produce a change in power configuration at either the local or national level.
Hypothesis 5: Competitiveness of elections has no effect on riots prior to democratic transition.
Since violence occurs between ethnic groups, it should follow that ethnically homogenous districts should be more peaceful than districts in which multiple groups reside. Since the proportion of ethnic groups relative to each other also determines coalitional calculations and expected outcomes,Footnote 28 the share of minorities relative to the rest of the population should also affect the likelihood of riots. The larger the proportion of minorities, the greater sway minority groups should have over politics. When the groups are relatively balanced, violence is more likely to occur.
Hypothesis 6a: Districts with a high level of ethnic fractionalization are more prone to violence than homogenous districts.
Hypothesis 6b: Districts with a minority share close to 50 per cent would have a greater likelihood of violence than districts with a very small proportion of minorities.
Hypothesis 6c: Districts with a larger share of minority population will have more violence than districts with a smaller share of minority population.
ELECTIONS AND COMPETITION IN INDONESIA
Indonesia is a good case with which to test these ideas. It transitioned to democracy after President Soeharto’s ouster in 1998, ending three decades of authoritarian rule. Since then, the country has decentralized, implemented direct elections of local executive leaders and held multiple national elections. As a new democracy, Indonesia shares many characteristics common among recently transitioned countries. The early years of democracy were difficult: it was recovering from the 1997 financial crisis, and immediately following Soeharto’s ouster, East Timor seceded and secessionist aspirations arose in other parts of the country.Footnote 29 It is currently a lower-middle-income country with a growth rate of 5 per cent, and has a medium level of human development.Footnote 30 Although the country is known as the world’s most populous Muslim country, Indonesia has over 300 ethnic groups and six officially acknowledged religions. Despite the multitude of ethnic groups in Indonesia, ethnic politics was largely muted in Soeharto’s time until the transition to democracy – at which point ethnocommunal violence and secessionist demands flared up – and has since declined.Footnote 31
Elections are not new in Indonesia; its first competitive election was in 1955. Considered to be the freest and fairest election in Indonesia’s history prior to 1998,Footnote 32 the 1955 election boasted an over 90 per cent turnout, and 172 parties/candidates competing for 272 seats in the People’s Representative Council and 560 seats in the Constituent Assembly. During Soeharto’s New Order regime, indirect elections occurred every five years in which voters voted for parties whose representatives would then elect a president. Soeharto forced the fusion of Muslim parties into Partai Persatuan Pembangunan (PPP) and secular nationalist and Christian parties into Partai Demokrasi Indonesia (PDI) in 1973, effectively reducing the number of parties to three: Golkar, PPP and PDI.Footnote 33
There was also little genuine competition. Although the laws afforded all three parties the same rights and responsibilities, PPP and PDI suffered many restrictions on their mobilizational capacity.Footnote 34 For instance, civil servants campaigned for Golkar under the guise of official duties well before the campaign period.Footnote 35 PPP and PDI raised their own funds, while Golkar had abundant financial resources amassed through the forced civil servants’ contributions to the Golkar foundation, Yayasan Dana Karya Abadi.Footnote 36 Golkar had party facilities and personnel across the archipelago down to the village level,Footnote 37 while PPP and PDI had little presence outside of Java. The official campaign period, shortened from sixty days in 1971 to twenty-five days in 1987 and 1997, further limited PPP and PDI’s ability to raise support. Moreover, PPP and PDI candidates were screened by New Order authorities, who would disproportionately disqualify them and pass those of Golkar.Footnote 38 Electoral fraud and violations of counting procedures were rampant and rarely penalized.Footnote 39 These created an uneven playing field for PPP and PDI and made genuine competition difficult.
While Golkar consistently dominated elections from 1971, opposition increased in the early 1990s. Megawati, daughter of former President Soekarno, was elected as a representative of PDI in 1987, and her popularity soared. When Soeharto replaced Megawati with Suryadi in the PDI leadership in 1996, hundreds of protesters took to the streets and violence ensued. Megawati and her party became a symbol of opposition to the regime.
Indonesia held its first democratic election in 1999. Forty-eight parties competed and five won the bulk of the vote. PDI-P won 33.7 per cent of the votes, while Golkar came second with 22.4 per cent and PKB, PPP and PAN (notably considered the country’s contemporary Islamic parties) won 12.6, 10.7 and 7.1 per cent, respectively.Footnote 40 In 2004, Indonesia held its first direct presidential election. Despite achieving second place in the 1999 election, Golkar garnered the majority of votes at 21.6 per cent.Footnote 41 Seventeen years after transition, Indonesia today has had three direct presidential elections (2004, 2009, and 2014), four legislative elections (1999, 2004, 2009, 2014) and numerous direct executive elections. In 2014, Indonesian voters elected Joko Widodo president, who narrowly defeated Prabowo Subianto who, because of his family connection to former President Soeharto and his involvement in violence during the New Order, represented for many a return to authoritarian ways.Footnote 42 Despite its rough start, Indonesia has shifted from autocracy to a multiparty, consolidated democracy.
DATA
To test the hypotheses outlined above, this article analyzes an unbalanced panel data of districts and municipalities in Indonesia from 1990 through 2005. The unit of analysis is district-year. The dataset includes a total of 5,371 district-years.Footnote 43 I chose this level of aggregation based on the arena of political contestation. A province-level analysis would be interesting, but it would collapse too many of the variations that occurred across districts.Footnote 44 Aggregating at the sub-district or village level would not identify the dynamics of competition that occur in the districts. Given the devolution of autonomy to the districts after decentralization, it is particularly important to understand how bids for power at the district level correlate with violence.Footnote 45
Riots
The dependent variable, Ethnic Riots, is measured as the count of reported riots in a district-year. I use data collected by Varshney, Tadjoeddin and Panggabean, who read local newspapers from fourteen provinces and compiled data on various types of communal violence from 1990 through 2003.Footnote 46 I focus only on their data on ‘ethnocommunal violence’. To extend the years of observation through 2005, I read the national newspaper Kompas (from 1999 through 2005) and the weekly magazine Tempo (from 2002 through 2005). Like Varshney and his co-authors, I assume that no-reports mean 0 incidents of violence occurred in a given district-year.
I also coded Riotsdeath, which documents casualty levels of each riot in a district-year, and Sevviol, an index for the severity of violence (from 0=no riots to 5=riots involving more than 500 deaths). These variables track both the level and intensity of violence in a district-year.
I found that 85.5 per cent of all the deaths in communal conflicts in Indonesia occurred in fifteen districts and municipalities, in which only 6.5 per cent of the country’s population lives.Footnote 47 Likewise, a yearly count of ethnic riots in Indonesia suggests that most riots occurred in the years immediately following the country’s democratic transition. This pattern suggests that the riots were both temporally and geographically clustered around particular years and districts.Footnote 48
Electoral Competitiveness
The main variable I use to measure electoral competitiveness is the margin of votes between first- and second-place winners in the district’s most recent parliamentary election.Footnote 49
The second indicator is Golkar Vote Share in the district’s most recent election. If political exclusion predicts riots, then it should follow that violence is concentrated in areas where barriers to entry are high, or in other words, where Golkar maintains strong control. I also coded variables that capture the reverse: the vote share of Golkar’s perceived opposition party, PDI-P, the vote share of Muslim parties combined,Footnote 50 and the vote share of all parties other than Golkar combined.
The last indicator captures a shift in Golkar’s popularity over the ten years prior to former President Soeharto’s ouster in 1998. It measures the percent change of Golkar vote share from 1987 to 1997. Positive values imply that Golkar performed significantly better in 1987 than in 1997, while negative values imply the opposite.
Turnover and Change in Electoral Competitiveness
To test whether the election of opposition candidates would reduce riots in post-conflict areas, I coded a dichotomous variable for electoral turnover in favor of non-Golkar candidates. I included another variable to measure change in electoral competitiveness: Delta Vote Margins. This variable is calculated as the difference in vote margins in the most recent election and those of the second-to-last election. Positive values in this variable imply an increase in competitiveness from the penultimate to the most recent election.
Temporal Order
Given the logic of my argument – that a lack of electoral competitiveness signals continued exclusion in the new government and motivates aspiring elites to mobilize their co-ethnics – it should follow that riots happen after uncompetitive elections, not before competitive ones. I created dummy variables for whether an observation falls during an election year, the year prior to an election or the year after an election. These variables provide insights into when unrest erupts in terms of proximity to the nearest election.Footnote 51
Ethnic Composition
Prior empirical research has typically used the Ethno Linguistic Fractionalization (ELF) index to measure ethnic diversity.Footnote 52 I added variables for religious and ethnic fractionalization, using district population data categorized by tribe and religion.Footnote 53 Given the co-ordination costs and ease of mobilization implied by group proportions in a diverse society,Footnote 54 I also measured the closeness of group balance to 50–50, calculated as the ratio of the share of the second-largest religious group over the share of the largest religious group in the district. The closer the ratio is to 1, the more evenly divided the district is between the two groups. An alternative indicator is included to capture the same concept: distance of the second-largest religious group’s proportion from 50 per cent. The smaller the distance, the closer the group balance is to 50–50.
Controls and Alternative Explanations
Some alternative mechanisms may be plausible. First, it is possible that low electoral competition does not capture the aspirations of voters on the ground. Rather, the prior regime party’s electoral dominance may be more indicative of electoral fraud, and anger due to suspicions of fraud may motivate violence.Footnote 55 My analysis does not test whether Golkar dominance influences riots through excluding opposition candidates or fraud. However, observers in both the 1999 and 2004 elections have noted that while there was evidence of vote buying and fraud in these first elections after 1998, they were not systemic, and these elections were considered a success.Footnote 56
Another alternative explanation for riots suggests that security capacity is an important factor in determining the onset of violence.Footnote 57 To account for this, I added a variable for district security spending (adjusted for inflation) based on figures reported in the annual district budget.
My argument focuses on local elites’ exclusion of opposition candidates and use of violence to attain political positions. However, elite manipulation is only one part of the story, since ordinary individuals can choose whether to riot.Footnote 58 Ordinary individuals may be motivated to riot due to poverty,Footnote 59 their engagement in social networksFootnote 60 or competition for jobs.Footnote 61 To account for the effects of poverty, I use logged GDP per capita.
In explaining Hindu-Muslim riots in India, Brass argues that prior riots institutionalize mechanisms that make clashes more likely to recur.Footnote 62 To control for the possibility of prior violence driving the results, I created a one-year lag variable of count of riots.
It is also important to account for regional effects on violence. Many observers have highlighted the differences in development levels between Java and the outer islands.Footnote 63 Furthermore, most incidents of ethnocommunal violence after Soeharto’s fall erupted outside Java, indicating that some regional effects may be at work.Footnote 64 To account for this, I created a dichotomous variable indicating whether a district is located in Java.
Temporal effects may also be at work. Many Indonesianists view riots as artifacts of the country’s transition to democracy, and assume that violence is largely a post-Soeharto phenomenon.Footnote 65 Furthermore, elections during the New Order were uncompetitive and often rigged, which could bias the results. I use a dummy variable indicating whether an observation is before or after Soeharto’s rule.
It has also been commonly suggested that riots are an urban phenomenon, and that proximity and frequent contact between groups leads to more clashes.Footnote 66 I added controls for district population density, logged population, logged area and a dummy variable for whether the unit is a municipality.
These factors aside, latent factors may have influenced ethnic riots that have not been accounted for by the indicators listed above. Among others, a district economy’s dependence on the stateFootnote 67 and the involvement of the security apparatusFootnote 68 are factors that may predict violence but for which I have no data at the moment. To control for district-specific characteristics that have been unaccounted for thus far, I also added district fixed effects in some of my estimations.
EMPIRICAL APPROACH
I conduct my analysis in three parts. First, I examine the relationship between electoral competition and ethnic riots in Indonesia. I use a negative binomial model because the distribution of the dependent variable fits this model better than a Poisson model.Footnote 69 Predictors of alternative measures for riots – riot-related deaths and severity of violence – are modeled with ordinary least square and ordered-logit models, as is common in the existing literature on violence.
Secondly, recognizing that the relationship between uncompetitive elections and the count of riots could imply either the losers instigating violence to leverage their positions or pro-Golkar supporters provoking clashes to intimidate potential challengers, I test whether districts that have become more competitive over time and experienced turnover (from Golkar to other parties) in subsequent elections are less prone to riots. If the exclusion logic explains outcomes in Indonesia, districts with turnover and increased competition would have less violence than districts without. This relationship is also modeled using a negative binomial model.
Thirdly, I highlight the dynamics of local elites’ competition for political positions and their roles in instigating violence in the first years after democratic transition in Poso, Central Sulawesi.
RESULTS
Electoral Competitiveness and Riots
The first three columns of Table 1 report the full sample results. Columns 4 and 5 report the results of regressions run only on Soeharto-era observations (1990–98) and post-Soeharto observations (1999–2005), respectively.Footnote 70
Table 1 Results
Note: Papua and Aceh districts were dropped in the models presented above. Controls are not shown due to space constraints; see the online appendices for the full results. ^p<0.10, *p<0.05, **p<0.01, ***p<0.001.
These results support my argument that uncompetitive districts tend to have more violence, even after accounting for district wealth, population, area, urban status, whether it is in Java and whether the observation is after Soeharto’s resignation in 1998.Footnote 71
Every unit increase in margin of votes is correlated with a decrease in differences in log of riot count of 0.01.Footnote 72 The incidence rate ratio (IRR)Footnote 73 of electoral competitiveness’ coefficient implies that with every 1 per cent increase in electoral competitiveness, the count of riots drops by 1 per cent.Footnote 74 The closer the most recent election was, the less likely the riot. Districts with wider vote margins, on the other hand, are more likely to experience riots. Estimates using alternative measures of riots – riot-related deaths and severity of violence – yield similar results.Footnote 75
Other measures of political competition also follow the outlined predictions. Golkar Vote Share is positively correlated with violence. Every 1 per cent increase in Golkar Vote Share leads to six times more riots in a district.Footnote 76 To test this idea further, I examine the relationship between particular party vote shares and riots. If uncompetitive districts dominated by Golkar have more violence, then the opposite should be true: districts in which Golkar’s rivals perform well would be more peaceful. The data show precisely this pattern. PDI-P vote share and the vote share of all parties other than Golkar combined are negatively correlated with riot counts.Footnote 77 Percent change in Golkar Vote Share in the last decade before Soeharto’s resignation is correlated with more riots. This result suggests that districts in which Golkar’s popularity had waned over the decade are more likely to experience riots than those in which Golkar’s popularity had increased or remained stable over time.
The Year After Election dummy variable also performs as predicted. Its IRR suggests that after-election years are twice as violent than other years. The year prior to an election, the year of election and proximity to election variables have no effect on riots.Footnote 78 This pattern supports the idea that violence in uncompetitive districts in Indonesia is an expression of frustration about recent election results rather than a voter mobilization strategy in the run-up to an election.
Another implication of my argument is that electoral competitiveness’ effects on riots should be noticeable only after 1998, since there was little prospect of change during authoritarian rule. To examine this, I controlled for whether observations were after 1998, and ran regressions on subsamples of the data. The results show that the After 1998 dummy variable is correlated with three times more violence. Before Soeharto resigned, the margin of votes has no effect on riots. Even though the size of the coefficient of vote margins in regressions of Soeharto district-years is larger than that of the coefficient of vote margins in regressions of after-1998 district-years, the former is statistically insignificant.Footnote 79
The composition of groups in a district predicts riots across various model specifications. The ratio of the second-largest religious group to the largest religious group in a district is positively correlated with the number of riots. The closer the group balance is to 50–50, the higher the likelihood of violence. Using alternative indicators to capture the effect of ethnic composition in a district on violence, I found that districts with higher religious fractionalization and districts in which the second-largest religious group’s proportion approaches half the population are more prone to violence.Footnote 80 These results suggest that excluded local notables utilize ethnic communities particularly when they are larger and thus presumably more useful for coalition purposes.
Thus far, I have shown that uncompetitive pro-Golkar districts tend to be more riotous than others. There is also evidence that the year following an election is more prone to violence than the year of or the year before an election. The data indicate that more riots occur in districts where the second-largest religious group is close to roughly half of the population than in districts where the second-largest religious group proportion is smaller. These patterns imply that the combination of low electoral competitiveness and a roughly balanced group composition in the years immediately following elections increases the likelihood of ethnic riots.
Two interpretations are possible. First, for Indonesia scholars who have argued that pro-Soeharto actors provoked violence to intimidate and assert their control, the correlation between riots and Golkar dominance provides evidence that violence is geographically concentrated in Golkar areas. The second plausible interpretation is that riots happen in Golkar-dominated areas because excluded actors mobilize violence to leverage their presence.
While Vote Margin is a good measure of electoral competition, and Golkar Vote Share accurately indicates the New Order regime entrenchment, neither of these variables directly captures ethnic political exclusion or local political configuration. Golkar is not an ethnic party, and neither is its perceived opposition party, PDI-P. The variable Golkar Vote Share in the prior election says nothing about which ethnic groups are represented in Golkar seats in parliament. It is possible, for example, that in one district, Golkar seats are occupied by members of certain ethnic groups, and by other ethnic groups in other districts. An ideal measure of ethnic political exclusion would require district-level data on the composition of the legislature and the legislators’ ethnic, religious and party affiliations, without which the pattern examined in this article would have limited information about why political exclusion produces ethnic riots. The turn from political to ethnic exclusion, rather, would have to be inferred from the presence of ethnic division in the district, as measured by ethnic composition in the district population.
It would be even better to have survey data of rioters that incorporate information on their ethnic and political affiliations, and their reasons for rioting. Such fine-grained information is not available for all districts in Indonesia for the fifteen years studied in this article. Absent these data, the best way to determine whether the proposed mechanism of exclusion holds is by testing whether an increase in turnover and electoral competitiveness reduces the likelihood of violence.
Turnover and Riots
If political inclusion were the driving motivation for violence, then riots would decline over time once non-Golkar candidates were elected. To test this, I examine how turnover in favor of non-Golkar parties and change in vote margins from one election cycle to the next correlate with the number of riots in various subsamples. An increase in electoral competitiveness and a greater representation of opposition candidates should be followed by a decline in violence, given the theory I outlined above.
Table 2 shows that turnover in favor of parties other than Golkar is correlated with lower incidents of riots. This finding is even stronger when the regression includes only uncompetitive districts.Footnote 81 The turnover variable loses its significance among uncompetitive, post-conflict districts after 1998,Footnote 82 which includes only fifty-eight observations. The higher the values in delta vote margins – the shift in electoral competitiveness from one election to the next – the lower the number of riots in a given district-year. These results suggest that violence dissipates when uncompetitive districts become more competitive and when more politicians affiliated with parties other than Golkar attain seats.
Table 2 Turnover and Riots
Note: Papua and Aceh districts were dropped in all regressions presented above. The dependent variable is count of riots. Controls are not shown. The full results are in the online appendices. ^p<0.10, *p<0.05, **p<0.01, ***p<0.001.
Taken together with the findings presented above, these results demonstrate support for the idea that in democratizing Indonesia, ethnic riots erupt in areas dominated by Golkar because continued Golkar popularity and low electoral competitiveness in the first elections after democratization signal continued exclusion at the local level. Once electoral competitiveness increases and turnover to opposition occurs, violence declines.
The Case of Poso, Central Sulawesi
Further evidence would require observations from more recent elections, since the period covered in this study stops at 2005 and thus includes only the first two elections after Indonesia transitioned to democracy. Since then, Indonesia has had two more legislative elections, in 2009 and 2014. These, along with Pilkada, the direct elections of local executive leaders that began in 2005, would provide good insight into whether elections have become more competitive, whether newcomers are placed into important political positions and whether these translate into a drop in violence in post-conflict areas. If the argument proposed in this article is correct, then new data from more recent years should show that ethnic riots decline in post-conflict areas that have become more competitive and have elected politicians affiliated with groups that were previously marginalized from politics.
Anecdotally, this pattern has been observed in Poso in Central Sulawesi.Footnote 83 Communal violence in Poso is one of the most complicated and protracted series of violence in Indonesia, starting in late 1998 and officially ending in July 2007. Claiming at least 900 lives, Poso violence evolved over time to include the intervention of external militias and terrorist groups in the later stages.Footnote 84 The early phases of violence were initiated by Christian and Muslim local elites’ desire to place their favored candidate in a district leadership position and their subsequent disappointment after Golkar appointed neither one of their championed candidates.Footnote 85
One Muslim community leader interviewed in Poso noted that ‘everyone was eagerly anticipating this election and who would replace Arief Patanga. Both [Christian and Muslim] communities had their championed persons, and both felt entitled to winning’.Footnote 86 In anticipation of the 1999 election, both Poso Christians and Muslims had their favored co-ethnic candidates, the Christian candidate backed by PDI-P and the Muslim candidate backed by PPP. Both candidates, however, were eliminated from the nominations due to their involvement in earlier riots in December 1998,Footnote 87 which lasted about a week. Golkar comfortably won in Poso in the June 1999 election and the party selected Muin Pusadan from outside the city as district chief, much to the dismay of local Poso elites from both Christian and Muslim communities.Footnote 88
The failure to place favored co-ethnic candidates through the 1999 election suggests that despite the country’s democratization, local political inclusion is not necessarily guaranteed. When Central Sulawesi Governor Paliudju later installed as district secretary an ethnic Bungku Muslim who neither the Poso Christian nor Muslim community supported, a prominent Poso community leader and district legislator from PPP released a statement of warning and threat: ‘If the people’s aspirations are neglected, that is, the aspiration calling for Drs. Damsyik Ladjalani to become Secretary of Poso, the riots of 1998 which had so afflicted this district will recur, and its scale will be even greater. This has been confirmed by several religious and community leaders in Poso’.Footnote 89 This statement was printed in the paper on 15 April 2000, and the second phase of Poso violence erupted the following day.Footnote 90 The timing of the statement’s release and the renewal of violence support the plausibility that local Poso elites used violence as a protest mechanism against disappointing political appointments. One Poso community leader I interviewed noted that, ‘After a period of relative calm and things have returned to normal, an incident would happen and conflict erupted again. There are people who gained from keeping Poso riotous’.Footnote 91
In the first Pilkada election in the district in 2005, an ethnic Pamona Christian, Piet Inkiriwang, and his running mate, Muthalib Rimi, won 42.6 per cent of the votes with a twenty-point lead over the second-place candidate. In 2010, he paired with Syamsuri and secured 38.76 per cent of the votes with a twelve-point margin. In 2005, Inkiriwang ran as a PDS-backed candidate, and in 2010 he ran with Partai Demokrat’s support. It is notable that not only did Inkiriwang seek the support of parties other than Golkar, he also strategically chose Muslim running mates both times. This practice of pairing candidates representing previously conflicting communities is prevalent in districts with relatively balanced ethnic groups.Footnote 92 This practice increases the political representation of previously marginalized groups that, in Poso’s case, also coincides with a decline in violence.
CONCLUSIONS
This article explores why ethnic riots erupted where and when they did in democratizing Indonesia. In the ethnic riots literature, the relationship between electoral competition and elite manipulation of ethnicity for electoral gains is well known. Most of what we know about ethnic riots, however, is based on clashes in mature democracies. Since most ethnic riots do not occur in mature democracies, it is important that we examine the pattern of riots in democratizing countries.
Using district-level data, I argue that ethnic riots tend to occur in ethnically divided districts with low electoral competition in the first years of democracy, because the lack of competition signals continued regime entrenchment at the local level despite democratization at the national level. Uncompetitive initial elections in new democracies portend regime entrenchment, half-hearted reform and the continued exclusion of opposition candidates. Consequently, uncompetitive elections – not competitive ones, as is the case in more mature democracies – would be more likely to produce riots in ethnically divided areas. This effect holds even after controlling for district-specific factors such as GDP per capita, ethnic diversity, prior levels of violence, national factors such as the end of Soeharto’s New Order regime, and regional factors such as whether a district is located in the outer islands or Java. In substantive terms, every 1 per cent increase in electoral competitiveness predicts a 1 per cent drop in the rate of riots. Every 1 per cent increase in Golkar vote share in a previous legislative election predicts six times more riots in the district. Considering that riots are relatively infrequent and often very deadly, this effect is noteworthy.
Also consistent with this article’s outlined implications, I find that riots generally occur after uncompetitive elections, not prior to competitive ones, as is the case for Hindu-Muslim riots in India. The data also provide evidence that turnover in favor of non-Golkar candidates and an increase in electoral competitiveness from one election to the next are correlated with fewer riots. As uncompetitive districts become more competitive, riots decline. These results suggest that riots in transitioning Indonesia are an alternative form of political engagement, in the face of disappointing election results that would not accommodate local elites’ demands for inclusion. Once these demands are met, violence declines.
These findings have several implications for research and policy. First, the results in this article support earlier findings that low competition in new democracies contributes to political instability.Footnote 93 Whereas this relationship has previously been observed across countries, this article demonstrates that the relationship also holds across regions within the same country. Secondly, this article connects two previously separate bodies of literature: it applies the exclusion logic used in examining armed rebellion to explaining ethnic riots, and it is the first to compare Wilkinson’s minimum winning coalition idea with the exclusion/inclusion argument offered by Cederman, Min and Wimmer. The findings also offer empirical evidence of the relationship between Golkar dominance and rioting in transitioning Indonesia.
While this article has highlighted the pattern between uncompetitive elections and ethnic riots during democratic transition, it leaves a number of questions unanswered. Due to the data limitations acknowledged earlier, it does not directly examine the impact of local ethnic political configuration on violence. It also is silent on who riots, what challengers and incumbents do before and after elections, and whether rioting has meaningfully altered the configuration of power in local governments. Answering these questions is essential for building a better theory of both election-related violence during democratic transition and of ethnic riots in general. The example from Poso district in Central Sulawesi suggests that the introduction of Pilkada, the direct election of local leaders, helped accommodate these demands for inclusion. Some scholars have noted that Pilkada candidates, particularly those who compete in post-conflict districts, have generally run alongside partners who represent an important ethnic minority group in the district – essentially a nod to the demands for inclusion.Footnote 94 In a future study I will examine whether this is a general pattern, and whether these arrangements were precisely what pacified clashes in post-conflict districts in Indonesia.
In terms of policy, these findings stress the importance of strengthening formal institutions that can channel and accommodate demands for inclusion in the early stages of democracy. In Indonesia, rioters took to the streets when formal institutions failed them, as Machado and others have suggested in their cross-national analyses.Footnote 95 Those who have benefited from the status quo will strive to protect their interests,Footnote 96 but this distributive tension would be played out more peaceably within formal political institutions. Designing institutions to encourage this outcome is challenging, but when institutions are not equipped to allow this tension to unfold and be resolved, the consequences are much worse.
Supplementary Material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S0007123415000423
