Content uploaded by Paul Kenny
Author content
All content in this area was uploaded by Paul Kenny on Jul 04, 2022
Content may be subject to copyright.
Content uploaded by Paul Kenny
Author content
All content in this area was uploaded by Paul Kenny on Dec 02, 2020
Content may be subject to copyright.
Commodity Booms, Conflict, and Organized Crime
Logics of Violence in Indonesia’s Oil Palm Plantation Economy
Paul D. Kenny∗
Rashesh Shrestha†
Edward Aspinall‡
July 1, 2022
Abstract
This paper examines the relationship between commodity booms and the incidence of group
conflict and criminality in the context of Indonesia’s expanding oil palm sector. Informed by
original fieldwork, the paper theorizes that violence will occur where there are choke points in a
commodity’s production supply chain. Such choke points create the opportunity and incentive
for armed groups – including “mafias”, youth gangs, landholders, and state security services –
to compete for control of an industry’s rents. Analyses of multiple sources of violent conflict
∗Professor, Institute for Humanities and Social Sciences, Australian Catholic University, 250 Victoria Parade, East
Melbourne VIC 3002, Australia; Visiting Fellow, Australian National University. Email: paul.kenny@acu.edu.au.
Corresponding author.
†Visiting Fellow, Australian National University. Email: rashesh.shrestha@gmail.com
‡Professor, Coral Bell School of Asia Pacific Affairs, Australian National University, 130 Garran Road, Acton ACT
2601, Australia. Email: edward.aspinall@anu.edu.au. This project was partly funded by the grant “Supporting the
Rules-Based Order in Southeast Asia (SEARBO), Australian Government, Department of Foreign Affairs and Trade
(DFAT), 2018—2021”, and by the Australian Research Council (ARC) Future Fellowship, grant no: FT120100742.
SEARBO is administered by the Coral Bell School of Asia Pacific Affairs at the Australian National University (ANU).
The views expressed in this paper are solely the authors’ and do not represent the opinions of DFAT, the ARC, or
the ANU. Ethics approval for fieldwork involving human subjects was obtained under ANU Human Ethics Protocol
2012/528 pursuant to grant no: FT120100742. Previous versions of this paper have been presented at the Associa-
tion for Asian Studies Annual Conference (Denver, 2019), University Padjadjaran (Bandung, 2019), the Australasian
Development Economics Workshop (Perth, 2019), the Annual Meeting of the American Political Science Association
(Washington, DC, 2019), and the Pacific International Politics Conference Online Speaker Series (2020). The paper
was circulated with this title as ACDE Working paper in trade and development, no. 2020/23. For comments on earlier
drafts we thank Nicholas Barnes, Lachlan McNamee, and Isamu Okada. We thank Johannes Pirker and colleagues for
sharing satellite data on oil palm coverage and suitability.
1
data at the individual, village, and sub-district levels support our main hypothesis that conflict
is related to the extent of oil palm production in a given area. We also find that the effect
of oil palm cultivation on conflict changes over time with prevailing commodity prices. Our
results are robust to the use of reduced form analysis to account for the potential endogeneity
of plantation expansion.
Key words: mafia; organized crime; violence; conflict; natural resources; oil palm
JEL Code: D74, L73, O13, Q33, Q34
2
1 Introduction
Commodity booms – large-scale increases in the demand for, and production of, raw materials or
primary agricultural products – are frequent occurrences. There is good reason to think that such
economic shocks could have disruptive social consequences including increased levels of predation
and conflict, especially in countries with weak institutional development (Collier & Goderis,2008).
However, recent empirical research indicates that not all commodity booms have the same effects.
Drawing on the classic “opportunity cost” model of Becker (1968), researchers have shown that
surges in demand for labor intensive agrarian commodities may increase incomes from agriculture
and hence depress conflict – an income effect – while elevated demand for raw materials may
increase inequality and the availability of lootable resources, thereby increasing the risk of violence
– a rapacity effect. (e.g. Abidoye & Cal`
ı,2021;Berman et al.,2017;Blair et al.,2020;Dal B´
o &
Dal B´
o,2011;McGuirk & Burke,2020;Dix-Carneiro et al.,2018;Draca et al.,2018;Dube &
Vargas,2013;Nillesen & Bulte,2014). Yet not all research confirms this agrarian commodity–
raw materials dichotomy (Bazzi & Blattman,2014;Fearon,2005;O’Brochta,2019). As inputs
to the narcotics trade, poppy and coca cultivation do not follow this pattern (Angrist & Kugler,
2008;Mejia & Restrepo,2013;Castillo et al.,2018;Bove & Elia,2013), while price increases in
a number of legal crops have also been found to be associated with elevated, not reduced, levels of
conflict (Crost & Felter,2019;Roett,2020;Madslien,2020).
Focusing on the expansion in production of oil palm in Indonesia over recent decades, this
paper provides further evidence for an association between agricultural commodity booms and in-
creases in low intensity conflict. Oil palm (elaeis guineensis) is an edible tropical oil, which grows
only within a narrow band of ten degrees north and south of the equator. Fuelled by an increase
in global demand and an attendant rise in prices in the mid-2000s (see Appendix Figure A1), In-
donesia rapidly expanded oil palm production from about 15 million tons in 2005 to 31 million
tons in 2015. In this period, exports surged (see Appendix Figure A2a), while the expansion of
the area under oil palm cultivation in Indonesia far outpaced that of any other estate crops (see
1
Appendix Figure A2b). The expansion took place in long standing producing areas in Sumatera as
well as new areas in Kalimantan and Papua, as shown in Appendix Figure A3. Because oil palm
is the highest income-earning crop among smallholder agribusinesses – a result we independently
corroborate – we might expect the boom to be associated with a drop in violent conflict. However,
qualitative data gathered from fieldwork in several palm producing villages of South Sumatra and
quantitative data at multiple levels of aggregation covering larger parts of Indonesia point in the
opposite direction.
To quantitatively estimate the relationships between oil palm production and violent con-
flict, we conduct econometric analyses at individual, village (desa), and sub-district (kecamatan)
levels. At the individual level, we conduct a primary survey of 1,920 individuals in nearly 200
villages across North and South Sumatra – two major oil palm producing regions within Indonesia
– in which we ask respondents about their experiences of violence and conflict in their village of
residence. For the village-level analysis, we utilize information on violence in the 2014 Village Po-
tential Survey (PODES). For the sub-district level analysis, we make use of the National Violence
Monitoring System (NVMS) dataset to study trends in various categories of violent conflict from
2005 to 2014. The categorization of violence and level of available detail varies across the datasets,
and the distinctions between different categories of violence are often blurry. Nontheless, at each
level of analysis (village or sub-district), we attempt to relate the occurrence of various types of
violence to the level of oil palm production in the village or sub-district and uncover characteristics
of violence associated with oil palm production.
The individual level survey results show that villages experience different rates of group
violence depending on the extent of oil palm production, with violence peaking at intermediate
levels of area under cultivation. The cross-sectional village-level analysis confirms that villages in
the mid-range of oil palm production have the greatest likelihood of experiencing inter-group vio-
lence. The dynamic analysis utilizing panel data at the sub-district level shows that the relationship
between the area under oil palm cultivation and conflict changes over time. The relationship be-
2
tween violent conflict and oil palm cultivation is stronger when international prices for oil palm are
high; resource conflicts escalate after 2008, peak in 2012, and gradually decline thereafter along
with commodity prices. Our quantitative results are robust to the potential endogeneity of oil palm
production, which we address by utilizing information on the agro-climatic suitability of a location
to oil palm cultivation (Edwards,2019).
We supplement our quantitative analysis with a qualitative investigation based on extended
primary fieldwork conducted in Indonesia in 2018-19, as well as by examining the description
of events recorded in the NVMS database. Drawing on this evidence, our interpretation of the
statistical results is that agrarian commodity booms are more likely to result in violence if and
where there are significant choke points in the supply chain of the commodity in question (Collier,
2000;Crost & Felter,2019). The production technology of oil palm yields multiple opportunities
for illicit gain by lightly armed rural criminal organizations, which we follow locals in calling “oil
palm mafias” (mafia sawit). The result is violent competition among various actors – including
not only such mafias, but also youth gangs (preman), local ethnic associations, and plantation
companies and their armed agents and allies – for control of the industry’s legal and illegal rents.
In general terms, the existence of production choke points, whether at the stage of harvesting,
processing, transit, or sale of a commodity, along with their amenability to large scale theft or
racketeering generates the incentive and opportunity for groups to engage in violent conflict.
This paper contributes to at least two related sets of research. First, it qualifies the notion that
booms in labor intensive agrarian commodities should be associated with a reduction in conflict.
Echoing the results of S´
anchez de la Sierra (2020), which examines the distinctive political effects
of gold and coltan mining in the Democratic Republic of the Congo, we argue that the particular
production characteristics of a commodity matter. Our quantitative results support those of Crost &
Felter (2019), which show that increases in the price of bananas is associated with conflict in banana
growing regions of the Philippines. Additionally, in our case, qualitative evidence from multi-
sited fieldwork indicates that the opportunity and incentives for armed group organization could
3
be a critical factor in linking agrarian commodity booms with violence. We show that like other
lootable natural resources, but unlike many bulk commodities, oil palm production is associated
with multiple choke points in the production process, which both facilitate extortion and foster
group rivalries for control over these illicit incomes. Our paper is thus close in spirit to that of
Dimico et al. (2017), which demonstrates that the Italian Cosa Nostra had its origins in the lemon
growing regions of Western Sicily during the nineteenth century boom in demand for citrus (as a
then recently discovered prophylactic for scurvy). In that case also, choke points in harvesting and
processing made extortion and theft by armed gangs or “mafias” possible, while the diffuse nature
of cultivation and absentee ownership made regular policing more difficult. Our evidence suggests
that a corollary to this kind of mafia predation is an increase in low-intensity conflict as armed
groups compete for a share of the industry’s profits.
Second, the fine-grained nature of our data, which includes various types of violence, allows
us to advance research on the effect of competition between armed groups for control over choke
points in both licit and illicit sectors of the economy. In South and Central America, the Sahel,
and elsewhere violence has been shown to be responsive to levels of competition between rival
mafias or gangs for control of the narcotics trade, especially in the context of demand and supply
shocks (Angrist & Kugler,2008;Castillo et al.,2018;Dur´
an-Mart´
ınez,2017;Magaloni et al.,2019;
Milla¸n-Quijano,2020;Yashar,2018). Our results indicate, however, that mafia or gang violence
does not depend on the illegality of the sector, or on the absence of the state as an enforcer of
property rights (Andreas & Wallman,2009;Yashar,2018). We find evidence that the rapacity effect
previously only associated with high value goods such as mineral oil or illicit commodities such
as cocaine, may also be present in at least some legal agricultural commodity sectors, where there
are potentially positive returns to racketeering and extortion (Gambetta & Reuter,1995;Reuter,
1987;Madslien,2020). Our quantitative results show a weak and inconsistent relationship between
commodity shocks and petty crime and interpersonal violence, thus supporting previous claims
that violence is being driven by organized groups or gangs, rather than individuals. With many
collective actors involved, some of them nominally legal, oil palm cultivation is associated with a
4
distinctive ecosystem of physical violence (Li,2017). The turn to cash crops such as oil palm may
provide both individual and collective economic benefits (Edwards,2019), but there is evidence
that it can also incur substantial social costs. We discuss the policy implications in the conclusion.
2 Theoretical Argument
Our theoretical understanding of the mechanisms linking oil palm plantation expansion to conflict
is informed by several months of original fieldwork conducted in 2018-19 in six randomly selected
oil palm producing villages in two districts in the province of South Sumatra, where 8.6 percent of
Indonesia’s oil palm was produced in 2016 (see Appendix Table A1). The two fieldwork districts,
Ogan Komering Ilir and Musi Rawas Utara, were first randomly selected from those districts which
had extensive oil palm plantation coverage and which had experienced violence in the recent past.
Cases were thus selected to be illustrative rather than representative (Seawright & Gerring,2008).1
Based on these observations, we theorize that level of oil palm production should be asso-
ciated with elevated levels of collective, if low-intensity, violence, as multiple armed entities fight
for control of the economic rents generated by the sector. All agricultural commodities, beyond the
quantity produced for subsistence, generate rents. These rents can be captured by primary produc-
ers, transporters, processors, the state, or indeed, any actor with the ability to use legal or illegal
means to monopolize the economic surplus generated at various points along the supply chain.
The total rent pool in a given sector is a product of extensivity (area under cultivation), intensity
1Fieldwork was carried out by one of the authors along with two Indonesia-based academics.
It included ethnographic observations gained from several weeks spent in the field along with ex-
tended interviews with multiple participants in the industry. Informants included members of local
communities, community leaders, plantation company staff, local police and government officials,
as well as NGO activists and academics. On the ground, we found that the prevalence of oil palm
plantation agricultural work was much more common in Ogan Komering Ilir, so focused follow-up
research on the villages in this sub-district. Further details on the interview procedure are presented
in Appendix A.
5
(yield), and market price. Agrarian commodity booms can have diverse effects on the likelihood
of violence depending on their associated production technologies, which in turn affect the level
and distribution of rents. The scale of production and the types of transportation and processing
required should impact the incentives for organized violent criminal activity, namely the extortion
of producers, transporters, and processors. The more that the production of a particular agricultural
commodity is dominated by large-scale plantations, and the greater degree to which there are choke
points in the transport, processing, and distribution stages of the supply chain, the greater are the
potential gains from theft and extortion relative to the cost (and risk) of that activity (Collier,2000;
Ornelas,2018).
The opportunity for predation in the oil palm sector is a result of the multiple physical
and temporal points in its production supply chain, which distinguish oil palm from that of many
other cash crops. Oil palm is largely monocropped, unlike some other cash crops, such as rubber,
which as a defoliating species can be co-planted with pineapple or other crops. Oil palm plantation
cultivation is also extensive, taking up vast tracts of territory. Although there is some variability in
yield – associated with plantation size and age –, increases in output in an administrative unit are
primarily driven by extensification (an increase in area under cultivation) rather than intensification
(higher yields per cultivated area). Thus, while data limitations mean that our empirical strategy
rests on an imperfect estimate oil palm production based on a measure of extensivity alone, in the
oil palm case this is a good approximation. The pool of rents to be captured varies mostly with
market prices and the area under cultivation.
In spite of its inherent extensivity, aspects of the oil palm production process can be highly
concentrated in both time and space. For large parts of the year, oil palm plantations can be lightly
populated, with a productive plantation needing only one worker for every 8-10 hectares compared
to one worker for every 2 hectares for rubber cultivation. At other times, however, activity on
plantations can be intense. Oil palm production requires the highly coordinated operation of mobile
work gangs to harvest the fresh fruit bunches (FFB), transport them by truck to processing mills,
6
and to spread fertiliser and pesticide through the plots. Because FFBs are heavy – weighing c.
25 kilograms each – and need to be taken for processing within 48 hours of being harvested, oil
palm production and harvesting is relatively capital intensive. The oil winning process involves the
reception of FFBs from the plantations, sterilizing and threshing of the bunches to free the palm
fruit, mashing the fruit, and pressing out the crude palm oil. The crude oil is further treated to
purify and dry it for storage and export. Even a medium-scale operation thus demands a substantial
processing and transport infrastructure, with processing concentrated into a small number of local
facilities. Moreover, because Southeast Asia has a single monsoon, it has only one peak-harvesting
season in contrast to Central and West Africa, which have two, further concentrating plantation
activity in time. This temporal and geographic concentration generate opportunities for predation
by organized armed groups.
The theft of bulky commodities such as oil palm fruit on an economically efficient scale, the
extortion of commodity transporters at road blocks, or the maintenance of a protection racket for
commodity processors are generally all beyond the capability of individual criminals. Moreover,
small-holder farmers may themselves be lightly armed, while larger plantations and processors can
afford to employ private security guards or engage the police or army for protection, increasing the
risk of theft to individual criminals. The main driving force behind inter-group violence is activity
associated with what local people call the “oil palm mafia”. The term “mafia” is used frequently in
Indonesia to refer to networks made up of corrupt officials, security officers, traders and criminal
gangs that specialize in extortion, theft, and manipulation in the markets for specific agricultural
commodities, with coffee, rice, shallot, sugar, tobacco, and maize “mafias” being among the other
examples. In sociological terms, a “mafia” refers to a specific type of criminal organization “which
produces, promotes, and sells private protection” (Gambetta,1996, 1). Our fieldwork revealed
that although some expropriation takes place in day time in the open (e.g., opportunistic theft
of harvested FFBs awaiting collection on roadsides), there is also a more organized version in
which gangs using trucks raid company estates at nights, rapidly and carelessly harvest FFBs, and
7
sell them to middlemen who then transport them to processing factories (often located in rival
plantations from where the theft occurred).
Strictly speaking, some of the entities in the Indonesian context – such as plantation secu-
rity forces – might be better described as “violent entrepreneurs” who use violence or the threat of
violence to gain an advantage in a market, especially where the state is weak or absent (Volkov,
2016). However, to the extent that these groups are themselves successful in establishing a local
monopoly of extortion on commodity production in a given territory, they often use their capacity
as specialists in violence to also engage in more prototypical “mafia” activity, such as protection
rackets and kidnapping. Indeed, officially registered plantation companies themselves often operate
in ways that violate formal laws or take advantage of legal “gray zones” (McCarthy,2004) (such as
overlapping land tenure and competing authority), and they frequently use state security forces (to
whom they provide private payments), security firms, local gangs and other militias and paramili-
taries to enforce their dubious legal claims. Assuming that gangs, mafias, and other agents operate
as rational, profit maximizing actors, we theorize that violent conflict results from competition
between armed groups over control of the rents generated in the sector.
Associated with this systematic mafia-style political economy is a general decrease in secu-
rity in oil palm producing areas. Interviewees repeatedly noted that villages in oil palm producing
districts are marked by high levels of violent predatory crime, including robbery and hold-ups on
highways and plantation roads, as well as kidnapping for ransom. Referring to one village with
this sort of violent reputation, residents of surrounding areas labelled it a “hell village” (kampung
neraka), while some village residents themselves took pride in their reputation for “hardness”. In
another instance, village youths ran a protection racket on the main road, in which they would pro-
vide “security” for oil palm and other goods trucks – requiring them to display stickers produced
by the group, and sometimes riding shotgun with drivers, accompanying them along stretches of
the road. In other locations, such as at government weigh stations, protection rackets are conducted
in collaboration with police and other state officials. In both of our case-study districts, local
8
land-rights advocates compiled many reports of legal disputes, protests and occupations, as well
as violent clashes between local landowners and plantation companies. To cite one of the most
deadly such clashes noted by our interviewees, a long-running land conflict in a village not far
from our case-study villages in Ogan Komering Ilir culminated in tit-for-tat violent clashes in April
2011: private security guards linked to the plantation company first attacked villagers, killing two,
prompting a reprisal attack in which villagers raided the company mess and killed five company
workers they encountered there. These and other observations suggest that the potential to control
illicit rents generates multi-sided conflicts between organized local and neighboring groups, with
possible spillover effects from village to village.
We also theorize that the violence associated with oil palm cultivation should be conditional
on prevailing commodity prices. Higher prices should increase competition over the rents, and
hence, violence. Because the number of armed groups is endogenous, it is empirically challenging
to estimate the causal effect of competition on violence directly. Following previous research we
posit that the extent of competition should be associated with demand (i.e., price) of a commod-
ity (Angrist & Kugler,2008;Castillo et al.,2018;Magaloni et al.,2019;Milla¸n-Quijano,2020;
Madslien,2020). Owing to the unusual characteristics of oil palm production that we described
above, in many respects it is closer to a capital-intensive resource than to a diffuse agricultural
one, implying that the rapacity effect may be stronger than the income effect. We thus anticipate
a price threshold across which violent competition among gangs, and between gangs and others in
the production chain (e.g., plantation companies, native landholders), should intensify.
Again this hypothesis finds tentative support in our fieldwork. Although we were not able to
directly observe long-term patterns of violence through our qualitative fieldwork – which occurred
at one point rather than over an extended period of rising and falling production – it is notable that
many of the large-scale instances of violence recounted by informants occurred during the post-
2005 boom. A possible explanation for this apparent violation of the opportunity cost hypothesis is
that even though farm gate prices may vary to a degree with international market fluctuations, the
9
income effect is likely to be small, thus having little impact on the cost of predation for individual
farmers or laborers. The biggest price shock is likely to be felt at the intermediary levels of transit
and processing, where profits and rents are concentrated. If the pool of rents increases sufficiently
due to higher prices, the potential prize for groups willing to violently contest control over the rents
associated with the transportation and processing of oil palm exceeds its costs. Consistent with this
understanding, interviewees explained that higher commodity price periods tended to reawaken
old resentments among dispossessed land owners, leading to renewed land claims and, especially,
increased confrontations and extortion between neighbouring smallholders, dispossessed villagers,
and mafias. The result was a higher incidence of group-scale violence.
3 Empirical analysis
In this section, we outline our approach to quantitatively examine the relationship between oil
palm production and different types of violence. The goal is to identify the types of violence
that are more prevalent in oil palm producing areas. The analysis is conducted at three levels:
individual, village, and sub-district. Each level of analysis provides a complementary picture of the
relationship between oil palm production and violence in Indonesia.
For the individual level analysis, we use a primary survey of 1,920 individuals conducted
in 2019 in 190 villages with varying levels of oil palm production. The focus is on the experience
and perception of group violence reported by the respondents. The village level analysis is done
using the 2014 Village Potential Survey (PODES) data, which provides information on occurrence
of conflict in the village as reported by village heads. The sub-district analysis uses the National
Violence Monitoring System (NVMS) database, which systematically collates violent incidents in
several Indonesian provinces for the 2005-2014 period. NVMS thus allows us to construct a panel
dataset in order to analyze the dynamics of violence resulting from the surge in oil palm prices over
10
this period. At each level of analysis we seek to distinguish between low-intensity but group-level
violent conflict and mere interpersonal violence or petty crime.
For our explanatory variable, ideally we would directly measure the total pool of rents or
profits locally produced by the oil palm sector. The rent pool will be primarily a product of both
the volume of oil palm produced (area x yield) and market price. For production levels, however,
because official data on production is given only at the district level, we have to estimate this
quantity indirectly for the village and sub-district. We estimate production using the area given
over to oil palm cultivation as a proportion of the total village area (for the individual and the
village-level analysis) or sub-district area (for the sub-district level analysis). Because oil palm is
planted in distinctive rows, the extensivity of production can be computed using remote-sensed data
that identifies pixels in satellite images (Austin et al.,2017). By overlaying the 2014 administrative
boundary map layer on the map identifying oil palm locations, we calculate the percentage area
under oil palm production at the village and sub-district level. Our oil palm data covers the regions
of Sumatra, Kalimantan, and Papua – where most oil palm production takes place, so our analyses
are conducted in villages and sub-districts in these regions.
Although the lack of data on yields makes our measure imperfect, as we previously noted, at
least in Indonesia, increases in output within administrative units are primarily driven by bringing
new areas under cultivation rather than by increasing yields. Production and area are thus strongly
related. To verify the accuracy of the satellite-based data, we check the correlation between our
oil palm intensity data and the official statistics reported by the Ministry of Agriculture, which
is available from the World Bank in its Indonesia Database for Policy and Economic Research
(INDO-DAPOER) database, at the district-level. We regress district-level oil palm production in
2010 on area under oil palm production calculated based on satellite data for various years and
report the coefficients and standard errors in Table A2. We also present a correlation graph in
Appendix Figure A4. Both results indicate high degree of correlation, indicating that the satellite
data is accurate. Another verification of the accuracy of our oil palm measurement comes from
11
independently sourced data on locations of oil palm mills and concession lands (Global Forest
Watch,n.d.). Villages with greater area under oil palm plantation coverage according to satellite
data are located closer to documented mills and concession lands. As shown in Appendix Table
A3, a 1 percent increase in distance from mills or concession land is associated with a 0.1 (10
percentage point) decrease in proportion of village area under oil palm cultivation. Thus we are
confident in the accuracy of our explanatory variable.
In our sub-district level analyses, we take into account potential issues with using oil palm
production extensivity as a causal variable. It is possible that some unobservable factors may
influence both the incidence of violence and oil palm production in a particular location. For
example, local property rights regimes or the strength of local institutions may be related to both
violence and new investments in the oil palm sector. These factors cannot be easily captured in the
data and could lead to bias in our estimates. Related to this, oil palm production may not expand
in areas with a very high risk of violence. Given the long-term planning required to reap rewards
from this particular commodity, investors may be hesitant to invest in areas that have pre-existing
risk factors that are likely to exacerbate violence. Due to this feedback effect from violent crime
to oil palm plantation expansion, the statistical relationship between the two variables could be
confounded.
The effect of time-constant unobserved factors can be taken into account in a fixed effects
regression model. Yet there may be other factors correlated with presence of oil palm that vary
over time and affect trends in violence. We account for this source of confounding by ensuring
that our estimates are robust to the inclusion of appropriate control variables and to the use of
reduced form estimation, with agro-climatic suitability for oil palm production instrumenting for
actual cultivation (Mejia & Restrepo,2013). Oil palm requires a certain climatic and geographic
conditions for it to be viable, including the slope of the land, rainfall patterns, and soil type (Pirker
et al.,2016). We do not expect the degree of palm oil suitability to be correlated with unobservable
influencers of violent crime, after controlling for observable characteristics of the location in the
12
baseline. Agro-climatic suitability is a particularly useful measure in this case, as it also partially
captures potential yields as well as potential area within a given administrative unit.
3.1 Individual level analysis
We use a probit model to analyse the factors that affect a respondent’s experience and perceptions
of violence. We concentrate on two outcome measures. First, the respondents were asked to what
extent the following problems examples of group violence occurred in the village: terrorism, reli-
gious conflict, ethnic conflict, and group conflict. They were also asked if other issues including
juvenile delinquency, thuggery, theft, other crimes, and conflict between village leaders and resi-
dents are a problem in the village. Second, the respondents were asked about violence and crime
the respondent believes occurred in the village over the past year. The various categories of violent
activity included in the survey questionnaire is listed in Appendix Table A4. The responses to the
questions are coded as 1 if the respondent answers in the affirmative and zero otherwise. In the
estimation, several related types of violence are combined to create composite dependent variables.
We model the probability that the indicator takes the value 1 conditional on oil palm production
intensity, controlling for other characteristics.
For each dependent variable, we estimate the following model:
violenceij =β0+β1low palmj+β2mid palmj+β3high palmj+αXij +γWj+eij (1)
where violenceij is individual i’s experience of violent conflict in location j,low palmj, mid palmj,
and high palmjare indicators of palm production extensivity in the village j. Villages with oil
palm area upto 20 percent are classified as low palm, 20-40 percent as mid palm, and above 40
percent as high palm. The comparison is made against villages that have no oil palm produc-
tion. Xij is a vector of individual level characteristics and Wjis a vector of village characteristics.
The village-level characteristics include the incidence of conflict and crime in 2014, the presence
13
of a police post within 5 km, and others. Individual control variables are a male dummy, age,
age-squared, education, income, economic activity (non-farmer, rice producer, coffee producer, oil
palm producer, rubber producer), and an urban residence dummy. For every dependent variable, in
the paper we report predictive margins of each village type. The predictive margin is the average
predicted likelihood of the dependent variable taking a positive value if all villages were of a given
type.
3.2 Village level analysis
For the village-level analysis, the measure of conflict includes an indicator variable for whether
amass fight was reported in the village over the previous year, as well as the number of such
mass fights. The incidence of such fights could occur amongst any groups in the village (residents,
students, tribes, or others) during the past year, with the underlying cause being the seizure of assets,
power, women, ideological/belief differences, sports, crowds/shows entertainment, or others. The
survey also further distinguishes between various types of mass fights based on the groups involved:
inter-community, intra-community, with security forces, with government officials, and amongst
students. The survey did not specifically ask about the reason behind the conflict. In addition, we
also have data on ten different types of crimes that provide evidence on state of the rule of law
in general, but are not related to group violence per se. The advantage of this dataset is that it
includes all villages in Indonesia, giving us a more complete picture compared to the individual-
level analysis (we have over 37,000 villages in the estimation sample for this level of analysis,
compared to 190 villages in the individual-level analysis). The main types of conflict and crime are
described in Appendix Table A5.
At the village level, we estimate the following regression model:
violencej=β0+β1low palmj+β2mid palmj+β3high palmj+αXj+ej(2)
14
where violencejis the measure of violent conflict in location j,low palmj, mid palmj,and
high palmjare indicators of palm production extensivity as defined above. Other relevant charac-
teristics of the village are accounted for by including control variables Xj. These control variables
include distance to district headquarters, distance to the closest oil palm processing mill, the per-
centage of households with electricity, whether the village is located within 5 km of a police station,
and whether the village is multi-ethnic. We also include district fixed effects, so that all comparison
is done among villages within the same district.
3.3 Sub-district level analysis
For the sub-district analysis, the dependent variable is a count of the number of violent conflict
events in each location. The NVMS database is created by recording incidents of violence re-
ported in local newspapers (National Violence Monitoring System (NVMS) dataset,2015).2Each
event is by classified by the data collection team into ten broad categories: “Resource conflict,”
“Governance conflict,” “Conflict of election and position,” “Conflict of identity,” “Popular justice,”
“Violence in law enforcement,” “Criminality,” “Domestic violence,” “Separatism,” and “Other.”
Several of the categories are further classified into sub-categories. The definition of each conflict
type is provided in Appendix Table A6. We conducted a content analysis of the different types of
violence, which we discuss below. We focus on Resource conflict and popular justice as the two
most relevant categories of inter-group violence for which we have panel data. Given that this data
is available for 2005-2014, we can study the dynamics of different types of violence, which is not
possible in the individual- and village-level analyses.
2Detailed description of the data compilation method is available in Barron et al. (2014).
15
To explore the interaction between expansion and commodity prices, we use the following
regression specification:
violenceit =α0+δpalmi+
2014
X
t=2006
βtpalmi×yeart+
2014
X
t=2006
θtXi0×yeart+
2014
X
t=2005
δtyeart+f ei+eit
(3)
Here, violenceit measures the incidence of violence in location iat time t,palmiis a measure of
extensivity of oil palm production in the location3,Xiis a vector of control variables measured
during the baseline year (2005), yeartis an indicator variable for year t,f eirepresents the fixed-
effects of location i, and eit incorporates other unmeasurable factors that may affect violent crime.
The estimates βt- one for each sample year - shows the trends in violent crime in areas with
varying intensity of palm oil cultivation, thus providing an impact of oil palm plantation expansion
on violent crime. Conceptually, this model tracks the evolution of violence in locations with varying
levels of oil palm production. Our specification mirrors the one used by Berman et al. (2017, pg.
1574), except that we use year dummies rather than prices of minerals in that year.4
We construct control variables from Pendataan Potensi Desa (PODES) and Census data.
The controls include those related to social characteristics (percentage of population that is Chris-
tian, percentage of migrants in population), economic characteristics (presence of plantation busi-
ness), political situation (voting patterns), and security apparatus (distance to nearest police station).
Except for the social characteristics, the controls are derived from PODES 2005, which collects in-
formation at the village level. To convert a village-level characteristic into those of the sub-district,
we calculate the share of households living in village with that characteristic. For example, with
3Since we are interested in the temporal pattern of conflict rather than differences across areas
at a point in time, we use a continuous measure of oil palm production extensivity rather than a
categorical one as in the individual- and village-level analyses. This allows us to focus on effect of
the year dummies.
4We only deal with single commodity - oil palm - rather than multiple minerals in Berman et al.
(2017).
16
distance to police station, we calculate the share of families living in villages where there is a police
station within 5 km.
4 Results
In this section, we present our findings from regression analysis of the impact of the area under oil
palm cultivation on different types of violent activity. We begin by providing a descriptive analysis
of the data, and then move to the discussion of the econometric analysis.
4.1 Individual level results
Summary statistics of basic respondent characteristics is given in Table A7 in the Appendix. The
average age of the respondents is 41 years. Education levels are higher in non-palm villages and
lowest in mid-palm villages. Income levels are higher in villages with the highest area of oil palm
coverage. Consistent with previous findings (Rist et al.,2010;Euler et al.,2017), our survey
confirms that oil palm farmers have higher incomes than other types of cash crop smallholders (see
Figure A7 in the Appendix). Oil palm farmers are more likely to be in the “high” income bracket of
self-reported average monthly household income. Almost half of palm farmers fall in this bracket,
compared to 30 percent of smallholders. A positive relationship between oil palm production and
violence would thus be inconsistent with the typical expectations of the income effect. Respondents
in high oil palm extensivity villages overwhelmingly Muslim. As expected, oil palm farming is the
most common occupation high oil-palm extensivity villages, although the share of rubber farmers
is also quite high.5
5Although the growth in rubber has not been as dramatic as oil palm, the presence of rubber
alongside oil palm complicates the interpretation of the relationship between conflict and oil palm
area. Rubber is also an important plantation crop in Indonesia and experienced a similar growth in
prices during the same period as oil palm. North and South Sumatra provinces also happen to be
17
Table 1reports the impact of oil palm production extensivity on perceptions of village prob-
lems reported by the respondents. We classify the different types of issues into five categories: all
group conflict (includes religious, ethnic, and group conflict), land and labour conflict, violence,
crime (includes juvenile delinquency, theft, thuggery, and other crimes), and other village issues
(includes problems with village leaders, corruption, and infrastructure). The dependent variable is
indicated in the column heading. We report the marginal effects of each level of oil palm produc-
tion, which are derived from Probit regression.
Table 1: Impact of oil palm production on village issues (marginal effects)
(1) (2) (3) (4) (5)
All group Land and Labour Violence Crime Village issues
Palm intensity
None 0.0433∗∗∗ 0.0910∗∗∗ 0.0902∗∗∗ 0.400∗∗∗ 0.365∗∗∗
(0.00805) (0.0117) (0.0105) (0.0197) (0.0193)
Low 0.0514∗∗∗ 0.0575∗∗∗ 0.118∗∗∗ 0.398∗∗∗ 0.249∗∗∗
(0.0114) (0.0123) (0.0174) (0.0261) (0.0232)
Mid 0.0701∗∗∗ 0.0841∗∗∗ 0.119∗∗∗ 0.326∗∗∗ 0.342∗∗∗
(0.0152) (0.0167) (0.0200) (0.0291) (0.0287)
High 0.0503∗∗∗ 0.0827∗∗∗ 0.102∗∗∗ 0.401∗∗∗ 0.359∗∗∗
(0.00956) (0.0119) (0.0131) (0.0215) (0.0208)
Observations 1917 1917 1917 1917 1917
Marginal effects; Standard errors in parentheses
* p<.1, ** p<.05, *** p<.01
Source: Authors’ calculation.
Note: Columns indicate different dependent variables: “All group” combines the categories of religious conflict, ethnic
conflict, and other conflict between groups; “crime” includes juvenile delinquency, theft, thuggery, and other crimes;
“Village issues” includes problems with village leaders, corruption, and infrastructure. Full list of village issues in-
cluded in the questionnaire can be found in Table A4. This table reports marginal effects derived from a Probit model;
full regression results can be found in Appendix Table A8.
Reports of group conflict as a problem are more likely in medium oil-palm production areas,
compared to non-producing areas (Column 1). While in non-oil palm villages the likelihood of
reporting a group conflict as a major problem is 4%, it increases to above 5% in oil palm villages,
with the highest rate of 7.5% in mid-level extensivity villages. Group conflict is reported to be an
issue with slightly greater likelihood in mid-level oil palm areas, but not in high oil palm extensivity
areas. Violence is also more likely to be reported by respondents in low- and mid-level oil palm
the main rubber producing locations. In our analysis of the survey data, we distinguish between
rubber producers and oil palm producers to add to the interpretation of our econometric results.
18
villages by 2.5 percentage points. The issues of crime, land and labour conflict, and village issues
are not significantly higher (and sometimes even lower) in oil palm producing areas. Perhaps
surprisingly, land and labor conflict are found to be a issue in all types of locations, not just in oil
palm producing ones. This is likely because rubber farmers, who are prevalent in oil palm areas,
also tend to report land conflicts at a high rate, as we found in a separate analysis of our data (not
reported).
The results above show that the occurrence of group conflict and violence was elevated in
oil palm producing areas, but do not tell us about the scale of the violence actually experienced.
To get at this, we use information in the survey on respondents’ experience of violence of differing
intensities. Table 2reports the estimated impact of oil palm production extensivity on low-, mid-,
and high intensity violence. The relevant survey questions inquired into some 20 categories of
violence and crime, which we organized into five categories: low-level violent conflict (physical
violence involving less than 3 people), mass physical violence involving 3 or more people, violent
crime (violence that leads to physical injury, including murder), other violence (including domestic
violence and sexual violence), and property crime (includes theft and robbery). The latter two
are included to exhaust all the options available in the survey questions – we have no theoretical
reasons to expect these two be salient in oil palm producing areas.
We find greater incidence of low-level physical violence and mass physical violence in oil
palm producing areas. The likelihood of physical violence being reported by the respondents in-
creases from 2.5% in no palm villages to 5.4% in mid-level oil palm villages. Similarly, the likeli-
hood of mass physical violence increases from 0.4% to 2.7% in mid-palm villages. The likelihood
of violent crime is also slightly higher in mid-level oil palm villages, although not in other levels of
extensivity. We do not find any higher likelihood of other violence or property crime being reported
being experienced in oil palm producing villages.
The overall conclusion from the individual-level analysis is that the occurrence of physi-
cal violence and group conflict are most prevalent in oil palm producing villages, with the most
19
Table 2: Crime in the village past year (predicted margins of palm extensivity)
(1) (2) (3) (4) (5)
Low violence Mass violence Violent crime Property crime Other crime
Palm intensity
None 0.0250∗∗∗ 0.00478∗0.0226∗∗∗ 0.585∗∗∗ 0.129∗∗∗
(0.00614) (0.00243) (0.00604) (0.0197) (0.0135)
Low 0.0111∗0.0130∗0.0164∗0.442∗∗∗ 0.137∗∗∗
(0.00558) (0.00540) (0.00704) (0.0266) (0.0182)
Mid 0.0544∗∗∗ 0.0286∗0.0376∗∗ 0.518∗∗∗ 0.0506∗∗∗
(0.0148) (0.0113) (0.0123) (0.0315) (0.0134)
High 0.00926∗0.0115∗0.00709∗0.524∗∗∗ 0.111∗∗∗
(0.00417) (0.00483) (0.00341) (0.0222) (0.0135)
Observations 1917 1917 1619 1917 1917
Marginal effects; Robust standard errors in parentheses
* p<.1, ** p<.05, *** p<.01
Source: Authors’ calculation.
Note: Dependent variables are indicated in column heading: ‘Low violence’ means physical violence involving 3 people or less; ‘Mass violence’
means mass physical violence involving more than 3 people; ‘Violent crime’ means violence that leads to physical injury and murder; ‘Property
crime’ includes theft or robbery; ‘Others’ includes domestic violence, sexual violence, and others. This table reports marginal effects derived from
a Probit model; full regression results can be found in Appendix Table A9.
elevated occurrences in mid-level oil producing villages. We did not find any impact of oil palm
production on most types of crime, or on conflict over land or labor per se.
4.2 Village level results
By way of summary statistics, Table A10 in the Appendix presents averages of selected village
characteristics by oil palm production extensivity. We see that palm producing villages show el-
evated levels of mass fights, with mid-level oil palm villages reporting the highest incidence and
frequency of mass mights. Intra-village conflict seems to be most salient in these villages. Crime,
theft, drugs, and gambling are also reported to be higher in oil palm producing villages compared
to non-producing villages. However, the summary table also makes clear that there exist other dif-
ferences between villages with and without oil palm production, such as number of families (palm
villages are larger on average, with mid-palm villages having the largest average population) dis-
tance to the district mayor’s office (palm villages are the further away), availability of police posts
20
(palm villages have lower likelihood of availability), and presence of several ethnic groups (palm
villages tend to be more diverse). These variables are used as controls in the regression model.
Table 3reports the village level estimation of equation 2. The dependent variables are
indicated in column heading. To enable comparison with the individual level results, we group
available types of conflict and crime into several variables. The first three columns pertain to
group conflict. The fourth column pertains to theft (which includes ordinary theft, violent theft,
and fraud/embezzlement), followed by interpersonal violence (persecution and murder), and other
crimes (burning, rape, drugs, gambling, and human trafficking). Unlike our primary survey, the
violence variable in this case is binary and does not provide detailed information on intensity or
type of violence. We use ordinary least squares to estimate the model and report the coefficients
on low-, mid-, and high-level oil palm production extensivity villages (no palm villages are the
reference group).
Table 3: Impact of oil palm extensivity on village conflict
(1) (2) (3) (4) (5) (6)
Num fights Intra-vill Inter-vill Theft Violence Others
Palm intensity (Ref: No palm)
Low 0.0143 0.00376 0.00464 0.0178 0.00301 0.0145
(0.00937) (0.00460) (0.00369) (0.0116) (0.00591) (0.0110)
Mid 0.0226** 0.0153* 0.00475 0.0275 0.00663 0.00160
(0.0102) (0.00834) (0.00399) (0.0178) (0.00739) (0.0130)
High 0.00975 -0.00307 0.00318 0.00599 -0.00870 -0.0264*
(0.00850) (0.00494) (0.00358) (0.0185) (0.00659) (0.0149)
Controls Yes Yes Yes Yes Yes Yes
Observations 37083 37083 37083 37083 37083 37083
Source: Authors’ calculation from PODES 2014.
Note: The table shows results from regressing incidence of fights and theft on village type. All regressions
include district fixed-effects, and columns 2 and 4 include additional control variables. Village boundaries
are based on 2014 definitions. Sample comprises of villages in Sumatra, Kalimantan, and Papua. Standard
errors clustered at the district level in parenthesis. * <.1 ** <.05 *** <.01.
We again find evidence of elevated group conflict in oil palm producing villages. Incidents
of fights are more likely to be reported in mid-coverage villages compared to no palm villages
21
(column 1). Mid-extensivity villages are likely to report 0.02 additional fights compared to no
palm villages. Given that the average number of fights is 0.03 (very few villages actually report
having mass fights), the coefficient is substantively meaningful. Most of the reported fights tend to
be intra-village, as indicated in column 2. The coefficients on violence are small and not statistically
significant, which is different from the results of the primary survey, where interpersonal violence
was found to be significantly elevated in oil palm producing areas. Although theft is most likely
to be reported in low- and mid-extensivity villages (column 4), the coefficients are not statistically
significant, which is consistent with property crime not being related to oil palm production in the
survey results.
4.3 Sub-district level results
4.3.1 Qualitative analysis of event descriptions
The NVMS contains 1,372 descriptions of violent incidents in which the Indonesian word for oil
palm (sawit) appears (excluding those in which the “sawit” appears as part of a place name). In
a proportion of these cases the oil palm appears to be only incidental to the violence (as when an
oil palm trader’s property is damaged in an attack on some other place of business or when there
is an incident of domestic violence in which a perpetrator works in the industry) and in some cases
(such as sexual assaults or dumping of bodies of murder victims) the main connection seems to
be that oil palm plantations are convenient places to perpetrate crimes because they are relatively
deserted. Early in the dataset, too, there is a number of reports of violent incidents in plantations
that are associated with the Free Aceh Movement insurgency which persisted in this northern part
of Indonesia until 2005.
The vast majority of cases, however, fall into one of two categories: incidents related to
robbery and incidents linked to company-community conflicts over land. With regards to robbery,
a great variety of violent incidents are described: robberies of oil palm traders at their places of
22
work, homes or while on the road; robberies of staff of plantation companies; robberies of oil palm
farmers or truck drivers transporting FFBs to, from or within plantations; theft from travellers in
vehicles passing through plantations but apparently unconnected to them; and theft of oil palm
FFBs, either while being transported, while in storage, or from fields. Such thefts could themselves
be accompanied by serious physical violence. Violence also often comes in the form of reprisals,
as when company security, or farmers, beat, torture or otherwise attack suspected thieves, or their
buyers, or when somebody accused of theft lashes out their accusers, or when the accused person’s
friends, family or neighbours engage in revenge attacks (e.g., by burning company property). Many
of the reported cases resulted in serious injuries or death, and were perpetrated with sharp weapons
such as machetes or cleavers, and, less frequently, firearms.
The second category – violence linked to clashes over control over land – was equally di-
verse. Violent incidents included large-scale clashes between villagers on the one side and company
security and/or security forces (often, the police mobile brigades, or brimob) on the other. Such
clashes could occur on disputed fields, such as when villagers tried to prevent company workers
accessing community land, or when private security and/or militia groups attacked farmers working
on their (disputed) fields. But they often also occurred on company premises, and involved attacks
by villagers on company offices, property, and equipment, often ending in the ransacking of offices
and burning or destruction of buildings and equipment. Reports of attacks by company forces on
villages, sometimes leading to burning of homes, were somewhat less common. There were also
numerous reports of deliberate cutting down, poisoning, or burning of seedlings or trees on com-
pany or private land, as well as clashes between protesters and security forces at demonstrations,
often at public buildings such as a local parliament or district head’s office. Violent incidents in
this broad category more typically involved large groups than instances of theft. The groups could
consist both of villagers (sometimes grouped into named organisations) and company security,
government troops, and private militias. However, it should be stressed that there was sometimes
considerable overlap in modes of violence and it was often difficult to disentangle theft motives
from disputes over land (e.g., as when large groups of villagers attack a company’s property when
23
one of their number has been accused of stealing from the company’s oil palm crop from company
land; or when a large group of villagers themselves raid company property or stop company trucks
in order to seize fruits grown on such land).
4.3.2 Descriptive statistics
Descriptive exploration of the data shows that the incidence of violence is different between oil
palm producing and non-producing sub-districts. The distribution of all violence types across oil
palm producing sub-districts is shown in Table A11 in the Appendix. We note that incidence of
resource conflict, popular justice, criminality, and domestic violence is much higher in sub-districts
with oil palm production in 2005 (column 2). A higher number of events per sub-district is recorded
in oil palm producing sub-districts. Among the four conflict types found to be more common in oil
producing sub-districts, domestic violence is most likely only indirectly related to oil palm activity,
so we focus on the other three for rest of the analysis. Resource conflict and popular justice more
clearly capture intra-group violence, while criminality includes instances of violence perpetrated
by individuals.
(a) Resource conflict (b) Popular justice (c) Criminality
Figure 1: Trend in various types of violence
Authors’ calculation from NVMS dataset. The figure plots coefficients on interaction between year
dummies and oil palm coverage by fitting a fixed-effects zero-inflated negative binomial distribution model
with incidence of each violence type as the dependent variable. Explanation of each conflict type is given in
Appendix Table A6.
We examine temporal patterns in these three types of violence. To visualize the pattern
of violence by type of location during 2005-2014 period, we run the baseline model specified in
24
Equation 3without any controls and plot the coefficients on interactions between year and the
area of a sub-district under oil palm cultivation. The coefficients are derived from fitting a fixed-
effects negative binomial distribution model on the violence data and are displayed in Figure 1. All
three types of violence show upward trends, but resource conflict and popular justice have stronger
trends than violent crime, as determined by the magnitude of the coefficients. The coefficients on
the interaction terms are larger in the case of resource conflict and popular justice. The peak of
resource conflict in 2012 coincides with the peak of commodity price boom observed in 2011.
4.3.3 Econometric results
The results from estimating the model represented by Equation 3with the full set of control vari-
ables are presented in Tables 4for resource conflict (Columns 1 and 2), popular justice (Columns
3 and 4), and criminality (Columns 5 and 6). Within each dependent variable, the first column re-
ports the result with no control variables, while the second column control for additional sub-district
characteristics (and their interactions with time indicators). To save space, only the coefficients on
the interactions between variable and year indicators are shown. The full set of controls used is
indicated at the bottom of the tables.
For the incidence of resource conflict, the coefficient on palm area are small and statisti-
cally insignificant. Thus, there was no baseline difference in incidence of resource conflict across
different levels of oil palm production extensivity. However, the coefficients on the interaction be-
tween palm area and years 2008, 2011, 2012, and 2013, with the highest magnitude for 2012, are
significant. This shows that the impact of palm production on incidence of resource conflict rose
over time. At its peak in 2012, a 10 percentage point larger oil palm area led to a 0.29 additional
incidence of resource conflict in a sub-district. Escalation of resource conflict coincides exactly
with the boom period for crude palm oil prices, implying that the economic value of the resource is
likely to be the main driver of violence. When adding control variables, the results are similar even
though the value of the coefficients are slightly changed.
25
Table 4: Estimation results on the impact of palm oil on various conflict - year interactions
(1) (2) (3) (4) (5) (6)
Resource Resource Pop. Jus. Pop. Jus. Criminality Criminality
Palm area -1.488 -0.662 0.556 0.995 1.330** 1.740***
(1.483) (1.096) (1.387) (1.249) (0.556) (0.593)
Palm area x 2006 0.447 0.426 -0.851 -0.872 -0.560*** -0.411*
(0.874) (0.973) (0.577) (0.568) (0.212) (0.237)
Palm area x 2007 1.327 1.389 -0.240 -0.143 -0.346* -0.347*
(1.156) (1.097) (0.591) (0.692) (0.180) (0.189)
Palm area x 2008 2.226*** 1.956** 0.609 0.737 0.184 0.259
(0.849) (0.913) (0.616) (0.639) (0.242) (0.221)
Palm area x 2009 1.847* 1.946* 0.454 0.452 -0.0695 -0.187
(1.089) (1.004) (0.711) (0.694) (0.211) (0.235)
Palm area x 2010 2.188** 2.198*** 0.881 0.796 0.0372 0.00630
(0.946) (0.801) (0.764) (0.607) (0.224) (0.264)
Palm area x 2011 3.110*** 2.872*** 0.761 0.849 0.0450 -0.00530
(0.865) (0.939) (0.716) (0.714) (0.303) (0.281)
Palm area x 2012 3.386*** 3.079*** 1.523** 1.515** -0.0194 -0.0257
(0.919) (0.768) (0.652) (0.690) (0.328) (0.286)
Palm area x 2013 3.002*** 2.531*** 1.492** 1.343** 0.689** 0.554*
(0.901) (0.861) (0.584) (0.567) (0.281) (0.290)
Palm area x 2014 2.222** 1.941*** 1.209* 0.992* -0.125 -0.304
(1.007) (0.729) (0.639) (0.527) (0.340) (0.402)
Social No Yes No Yes No Yes
Security No Yes No Yes No Yes
Election No Yes No Yes No Yes
N12134 11377 16655 15619 22440 20868
Bootstrapped standard errors in parenthesis.
*<.1 **<.05 ***<.01
Source: Authors’ calculation. The table show results from negative binomial fixed-effects regression with various
indicators of conflict as dependent variable and palm area in 2015 interacted with year indicator as regressor of interest.
Columns 1 and 2 show results for resource conflict, columns 3 and 4 show results for criminality, and columns 5 and
6 show results for popular justice. Within each depend variable, the first column only includes the main regressors,
while the second column includes the following additional control variables: sub-district’s share of Christians, share
of migrants, share of families living in villages within 5km of a police station, share of families living in villages that
voted for Golkar and PDP. Each of the additional control variables is also interacted with year dummies. Sub-districts
are defined based on their 2000 boundaries.
In the NVMS data, one subcategory under resource conflict is land conflict. Recall that
in earlier analysis at the individual-level we did not find differences in incidence of land conflict
across villages with different levels of oil palm production at one point in time. However, in the
present dynamic analysis, we do find that oil palm areas experienced an escalation in land conflict
over time. This is confirmed by plotting the number of incidences of land conflict across time by
26
level of oil palm production, as is done in Appendix Figure A8. The number of reported incidents
of land conflict rises along with oil palm boom in locations that have oil palm activity.
On popular justice, we see no differential incidence of popular justice until 2011, after
which there was a sharp escalation during the 2012-2014 period in sub-districts with greater oil
palm production extensivity. A relatively higher incidence was recorded in 2013, when a sub-
district with one percentage point larger area in oil palm recorded 1.6 additional popular justice
incidence, as indicated by the coefficient on the interaction term between oil palm and year 2013.
The results indicate a surge in group conflict around time when the value of oil palm rose.
On the incidence of criminality, the coefficient on palm area is large and statistically signif-
icant. This implies that in general, sub-districts with a higher levels of oil palm production had a
greater incidence of criminal activity. A ten-percentage point increase in oil palm area is associated
with a 0.13 additional incidence of criminality. The incidence of criminality is falling over time
as indicated by the negative coefficients on year interactions, but the impact remains net positive.
For instance, in 2012, a sub-district with ten percentage point greater coverage of oil palm saw
0.2 [=(2.53 - 0.51)/10] additional incidents of criminal activity. The main results are robust to the
inclusion of additional control variables. In sum, the incidence criminality is higher in areas with
oil palm production, although it does not increase over time with oil palm prices. Although earlier
results did not find cross-sectional differences between incidence of theft across villages with dif-
ferent levels of oil palm production, temporal analysis shows that incidence of one subcategory of
popular justice - retaliation over theft - increased significantly in areas with oil palm production,
although the levels were higher in non-producing areas compared to low palm and high palm areas.
This is depicted in Appendix Figure A9. Thus, echoing the insights from our fieldwork, the surge in
oil palm prices does appear to have specifically induced theft of crops, which in turn led to violent
group retaliation.
27
4.3.4 Robustness checks using oil palm suitability
To ensure that the above results are not confounded by the endogeneity of oil palm production, we
make use of information on sub-districts’ suitability for oil palm production based on a composite
index of soil type, topography, rainfall, temperature, and existing land usage (Pirker et al.,2016).
A sub-district’s suitability for oil palm production is a natural endowment that is not affected by
relevant political or economic factors. Thus, this variable is plausibly exogenous to the outcome we
are interested in explaining. Furthermore, we can be confident that the direction of causality runs
from suitability to oil palm production extensivity (Edwards,2019). Likewise, it is plausible that
suitability does not directly impact any other drivers of violent conflict, other than through its effect
on oil palm production, thus meeting the exclusion restriction. The possible exception we note is
state presence, which we control for in additional specifications. Suitability should be related to
both the intensity and extensivity of production making it a good candidate instrument to use in an
instrumental variables approach to correct for the endogeneity of oil palm production.
First, we check the correlation between palm area coverage in 2015 and the suitability of a
sub-district for oil palm cultivation. To construct the suitability variable, each pixel of a subdistrict
is classified into six levels of suitability coded from zero (lowest) to five (highest) (see Figure A6 in
the Appendix). We then compute the proportion of subdistrict area with various levels of suitability.
Among sub-districts in our estimation sample, on average, 28 percent of total sub-district area is
under the lowest suitability level while 9 percent is under highest suitability level. As expected, oil
palm suitability is highest in Sumatra, Kalimantan and Papua. Table A12 in the Appendix shows
the results of regressing oil palm area in 2015 and 2005 on the proportion of subdistrict area under
various levels of suitability. The coefficients all have the expected sign – having more area under
highest level of suitability increases a sub-district’s area under oil palm production, while having
more area under level 1 suitability reduces a sub-district’s area under oil palm cultivation. Thus
suitability is a strong predictor of actual area under oil palm cultivation.
28
Our preferred approach is to use the suitability variable to replace oil palm production as a
regressor in estimating Equation 3to get reduced form results. Our theoretical arguments are based
on the output of raw and lightly processed oil palm in a given location, which creates opportunities
for predation and thus fosters group conflict. However, our main explanatory variable is based on
satellite data, which identifies areas of oil palm plantation by categorizing pixels in photographs
taken from space and thus captures spatial coverage of oil palm trees. While extensivity is corre-
lated with production, GIS data is necessarily an imperfect proxy of actual output levels. Because
we lack data on production below the district level, instrumental variables estimation in which pre-
dicted values from a first stage regression of oil palm extensivity on suitability could thus introduce
further noise to an already complex analysis. The causal effect of suitability (which instruments for
both extensivity and intensity of production) on violence can be more easily interpreted.
We can interpret the result of this new estimation as a study of trends in conflict in sub-
districts with various levels of suitability for oil palm cultivation. To keep the model tractable, we
construct a single suitability variable that is simply the total sub-district area under the highest two
levels (4 and 5) of suitability. The models otherwise remain the same. The result for resource con-
flict is presented in the Appendix in Table A13. The structure of this table is similar to Table 4. The
results are qualitatively similar even though the coefficient estimates are different. Sub-districts
where a larger share of suitable area saw a growing intensity of resource violence between 2009
and 2013. Similar to earlier results, the peak impact is in 2012. The results in the case of popular
justice also lead to a similar conclusion. The incidence of popular justice increased significantly
in high oil palm suitability sub-districts during the 2012-2014 period. When using criminality as
the dependent variable, the coefficient on suitability is positive and statistically significant. Fur-
thermore, we find that the impact of suitability on the incidence of criminality is actually higher in
2013, similar to the main results.
Although the general conclusion from the suitability result is also that areas with more palm
production tend to have greater incidence of criminality, the dynamic pattern is different from our
29
baseline model. The estimates are smaller than those in Table 4because the suitability measure
does not predict perfectly the actual oil palm production in a sub-district.
5 Discussion and conclusion
This paper examines the evolution in violence in response to the surge in production of oil palm in
Indonesia over the last two decades. Our quantitative analysis yields evidence of distinct patterns of
violence in oil palm producing areas. We find that areas at the mid-level of extensivity of oil palm
coverage experience an elevated incidence of group conflict and violence, although not consistently
so of crime in general. We also find that the surge in export value of oil palm led to a rise in the
incidence of two key forms of collective violence – resource conflict and popular justice – over
time in areas with a larger area under oil palm production.
Although the relationship between prices and violence has recently been observed with
respect to other commodities, the non-linear relationship between plantation coverage and violence
appears to be new. We propose two possible explanations for this relationship. The first is due to
low population densities in villages with large oil palm intensities, where the largest commercial
plantations exist. Conflict would theoretically seem to be more likely where population densities,
and hence group rivalries, are greater. The second explanation rests on the idea that areas with
the highest levels of plantation coverage are also likely to be the most established. In contrast,
in areas with relatively little coverage to begin with, the potential for growth in plantation area is
large, drawing in new groups to compete over the rents. We can see this in the data by plotting the
relationship between new oil palm area since 2005 and percentage of village area under oil palm
cultivation. Figure A10 in the Appendix shows that there is an inverse-U relationship between
gaining new oil palm cultivation and oil palm coverage in 2015. The largest gains in oil palm area
are observed in villages that are in the mid-level oil palm. Thus, these fast growing areas may be
more prone to conflict.
30
Our empirical methodology extends the work of recent research in political science and eco-
nomics that has attempted to estimate the causal effect of resources on conflict using sub-national
data on the location of resources or crop production and conflict. The granularity of our data, how-
ever, allows us to go beyond existing research in at least three crucial respects. First, covering both
non-fatal violence in addition to larger-scale events, our data allow us to probe low intensity out-
comes that are unobservable in most other studies. Second, with multiple data sources at different
levels of analysis, including an original survey, we can have greater confidence in the reliability of
our results than those resting on a single data source. Third, because oil palm requires very specific
growing conditions, we are able to employ reduced form estimation strategies to deal with possible
confounding in the location of plantation expansion.
Our qualitative investigation also provides important insights on the causal mechanisms
at work. Previous research could largely only speculate on the role that armed groups played
in accounting for the correlation between commodity prices and violence. Data gleaned from
interviews and ethnographic fieldwork strongly indicated that the violence observed in and around
oil palm plantations is primarily driven by lightly armed groups rather than individuals. Predation in
the sector is the result of organized criminal activity, rather than being a mere crime of opportunity.
Our paper suggests that future research on the effect of commodity price shocks on crime
and conflict needs to be wary of over generalization. It seems likely that the ambiguous aggre-
gate effects of commodity price movements on violence observed in some studies could be due
to multiple countervailing effects, rather than the strict absence of any effect at all. Not only are
there differences between the raw materials and agrarian commodity sectors, but also within these
sectors. The oil palm industry, for one, possesses many of the features of high-value resources and
illegal commodities with a rapacity effect predominating over an income effect.
Our results have ambiguous implications for policy. On the one hand, our survey results
confirm that higher incomes can be obtained by farmers in high demand export crop industries.
Moreover, the benefits may go beyond those directly engaged in the sector. As others have shown,
31
oil palm production has been associated with positive economic spillovers Edwards (2019). In
Indonesia, even as 5.9 million people are directly employed in the sector, an additional 50 million
are indirectly dependent on it. The country is now the world’s largest exporter. However, this
dependence can foster problems. The expansion of the area under oil palm cultivation has been
associated not only with environmental problems, but also, as we have shown increased social
dislocation in the form of predation and violence. In the Indonesian case, it remains to be seen
whether the level of governance can catch up with the expansion in area under cash crop production.
References
Abidoye, Babatunde, & Cal`
ı, Massimiliano. 2021. Income Shocks and Conflict: Evidence from
Nigeria. Journal of African Economies,30(5), 478–507.
Andreas, Peter, & Wallman, Joel. 2009. Illicit markets and violence: what is the relationship?
Crime, Law and Social Change,52(3), 225–229.
Angrist, Joshua D, & Kugler, Adriana D. 2008. Rural windfall or a new resource curse? Coca,
income, and civil conflict in Colombia. The Review of Economics and Statistics,90(2), 191–
215.
Austin, KG, Mosnier, A, Pirker, J, McCallum, I, Fritz, Steffen, & Kasibhatla, PS. 2017. Shifting
patterns of oil palm driven deforestation in Indonesia and implications for zero-deforestation
commitments. Land Use Policy,69, 41–48.
Barron, Patric, Jaffrey, Sana, & Varshney, Ashutosh. 2014. How large Conflict Subside: Evidence
from Indonesia. Indonesian Social Development Papers.
Bazzi, Samuel, & Blattman, Christopher. 2014. Economic shocks and conflict: Evidence from
commodity prices. American Economic Journal: Macroeconomics,6(4), 1–38.
32
Becker, Gary S. 1968. Crime and punishment: An economic approach. Journal of Political Econ-
omy,76(2), 169–217.
Berman, Nicolas, Couttenier, Mathieu, Rohner, Dominic, & Thoenig, Mathias. 2017. This mine is
mine! How minerals fuel conflicts in Africa. American Economic Review,107(6), 1564–1610.
Blair, Graeme, Christensen, Darin, & Rudkin, Aaron. 2020. Do Commodity Price Shocks Cause
Armed Conflict? A Meta-Analysis of Natural Experiments. American Political Science Review.
Bove, Vincenzo, & Elia, Leandro. 2013. Drugs and violence in Afghanistan: A panel VAR with
unobserved common factor analysis. Defence and Peace Economics,24(6), 535–554.
Castillo, Juan Camilo, Mej´
ıa, Daniel, & Restrepo, Pascual. 2018. Scarcity without leviathan: The
violent effects of cocaine supply shortages in the mexican drug war. Review of Economics and
Statistics.
Collier, Paul. 2000. Rebellion as a quasi-criminal activity. Journal of Conflict resolution,44(6),
839–853.
Collier, Paul, & Goderis, Benedikt. 2008. Commodity prices, growth, and the natural resource
curse: reconciling a conundrum. Growth, and the Natural Resource Curse: Reconciling a Co-
nundrum (June 5, 2008).
Crost, Benjamin, & Felter, Joseph H. 2019. Export Crops and Civil Conflict. Journal of the
European Economic Association,18(3), 1484–1520.
Dal B´
o, Ernesto, & Dal B´
o, Pedro. 2011. Workers, warriors, and criminals: social conflict in
general equilibrium. Journal of the European Economic Association,9(4), 646–677.
Dimico, Arcangelo, Isopi, Alessia, & Olsson, Ola. 2017. Origins of the sicilian mafia: The market
for lemons. The Journal of Economic History,77(4), 1083–1115.
Dix-Carneiro, Rafael, Soares, Rodrigo R, & Ulyssea, Gabriel. 2018. Economic shocks and crime:
Evidence from the brazilian trade liberalization. American Economic Journal: Applied Eco-
33
nomics,10(4), 158–95.
Draca, Mirko, Koutmeridis, Theodore, & Machin, Stephen. 2018. The changing returns to crime:
do criminals respond to prices? The Review of Economic Studies,86(3), 1228–1257.
Dube, Oeindrila, & Vargas, Juan F. 2013. Commodity price shocks and civil conflict: Evidence
from Colombia. The Review of Economic Studies,80(4), 1384–1421.
Dur´
an-Mart´
ınez, Ang´
elica. 2017. The politics of drug violence: Criminals, cops and politicians in
Colombia and Mexico. Oxford University Press.
Edwards, Ryan B. 2019. Spillovers from agricultural processing. Work. Pap., Dartmouth Coll.,
Hanover, NH.
Euler, Michael, Krishna, Vijesh, Schwarze, Stefan, Siregar, Hermanto, & Qaim, Matin. 2017. Oil
palm adoption, household welfare, and nutrition among smallholder farmers in Indonesia. World
Development,93, 219–235.
Fearon, James D. 2005. Primary commodity exports and civil war. Journal of conflict Resolution,
49(4), 483–507.
Gambetta, Diego. 1996. The Sicilian Mafia: the business of private protection. Harvard University
Press.
Gambetta, Diego, & Reuter, Peter. 1995. Conspiracy among the many: the mafia in legitimate
industries. Pages 99–120 of: The Economic Dimensions of Crime. Springer.
Global Forest Watch. Oil palm concessions.
Li, Tania Murray. 2017. After the land grab: Infrastructural violence and the “Mafia System” in
Indonesia’s oil palm plantation zones. Geoforum.
Madslien, Simen. 2020. Are avocados the blood diamonds of Mexico?: an empirical study on how
increasing demand for Mexican avocados is related to cartel violence.
34
Magaloni, Beatriz, Robles, Gustavo, Matanock, Aila M, Diaz-Cayeros, Alberto, & Romero, Vidal.
2019. Living in Fear: the Dynamics of extortion in Mexico’s Drug War. Comparative Political
Studies, 0010414019879958.
McCarthy, John F. 2004. Changing to gray: decentralization and the emergence of volatile socio-
legal configurations in Central Kalimantan, Indonesia. World Development,32(7), 1199–1223.
McGuirk, Eoin, & Burke, Marshall. 2020. The Economic Origins of Conflict in Africa. Journal of
Political Economy,128(10), 3940–3997.
Mejia, Daniel, & Restrepo, Pascual. 2013. Bushes and bullets: illegal cocaine markets and violence
in Colombia. Documento CEDE.
Milla¸n-Quijano, Jaime. 2020. Internal Cocaine Trafficking and Armed Violence in Colombia.
Economic Inquiry,58(2), 624–641.
National Violence Monitoring System (NVMS) dataset. 2015. Government of Indonesia / World
Bank.
Nillesen, Eleonora, & Bulte, Erwin. 2014. Natural Resources and Violent Conflict. Annual Review
of Resource Economics,6(1), 69–83.
O’Brochta, William. 2019. A meta-analysis of natural resources and conflict. Research & Politics,
6(1).
Ornelas, Ruth G. 2018. Organized crime in Michoac´
an: Rent-seeking activities in the avocado
export market. Politics & Policy,46(5), 759–789.
Pirker, Johannes, Mosnier, Aline, Kraxner, Florian, Havl´
ık, Petr, & Obersteiner, Michael. 2016.
What are the limits to oil palm expansion? Global Environmental Change,40, 73–81.
Reuter, Peter H. 1987. Racketeering in legitimate industries. Rand Corporation.
Rist, Lucy, Feintrenie, Laur`
ene, & Levang, Patrice. 2010. The livelihood impacts of oil palm:
smallholders in Indonesia. Biodiversity and Conservation,19(4), 1009–1024.
35
Roett, Katie. 2020. Green Gold: Avocado price shocks and violence in Mexico.
Seawright, Jason, & Gerring, John. 2008. Case selection techniques in case study research: A
menu of qualitative and quantitative options. Political research quarterly,61(2), 294–308.
S´
anchez de la Sierra, Ra´
ul. 2020. On the Origins of the State: Stationary Bandits and Taxation in
Eastern Congo. Journal of Political Economy,128(1), 32–74.
Volkov, Vadim. 2016. Violent entrepreneurs: The use of force in the making of Russian capitalism.
Ithaca: Cornell University Press.
Yashar, Deborah J. 2018. Homicidal Ecologies: Illicit Economies and Complicit States in Latin
America. Cambridge University Press.
36
A Appendix: Interview Procedure
1. Language: All the interviews were conducted in the Indonesian language by Indonesian-fluent re-
searchers, including one of the authors and two research assistants.
2. Location: Interviews were conducted in a variety of safe locations in selected fieldsites – village
offices, village halls, local foodstalls, etc.
3. Sampling: We used snowball sampling methods, first seeking information about informed persons in
each case-study village from contacts in the regional capital (academics, NGO activists, lawyers, etc)
and then from village heads, officials and informal community leaders, as well as from persons we
encountered in our daily interactions in the villages.
4. Confidentiality: We promised our interviewees confidentiality (so long as the law allows) and main-
tained confidentiality by a.) not recording interviews (experience tells us that interviewees in rural
Indonesia often fear that recordings may fall into the wrong hands); b.) using identifying codes rather
than interviewee names on notes; and c.) not disclosing information provided by any interviewee to
any other person during our field visits (i.e. we did not mention names of interviewees when triangu-
lating/confirming material). We also explained we would not name our interviewees in any research
outputs, unless they wanted to be named (in our experience public figures sometimes prefer to be
named on matters of public record; in this research no interviewee chose that option).
5. Informed consent: Oral consent was obtained by reading out a brief consent script at the start of each
interview, and asking the interviewees at the end whether they consented to be interviewed. Written
consent forms are not appropriate in the Indonesian rural context (being largely associated with offi-
cialdom and with many interviewees having limited literacy). The oral consent script summarized the
background of the research, explained the topic of the interview and how long it would take, explained
that we would maintain confidentiality so long as the law allowed, explained how we would store the
data, explained that we would not name them in research outputs unless they wanted to be named
(and asked for their choice), told them they could halt the interview and withdraw from the research
at any time, and that they could refuse to answer any question without giving any reason. At the end
of the script, we paused to ask whether they had understood, and whether they had questions. If all
was clear, we then asked if they consented to the interview.
6. We provided interviewees with information on how to contact the researchers and/or the appropriate
university ethics board should they wish to follow up on the research or had problems with the conduct
of the research.
Back to Section 2.
37
B Appendix Tables
38
Table A1: Palm production area (thousand hectares) by provinces, 2011 – 2016
2011 2014 2016 % Change
2011 to 2014
Province share
2016(% of na-
tional produc-
tion)
Aceh 360.2 420.2 441.3 16.7 3.7
Sumatera Utara 1164 1396.3 1445.7 20 12.13
Sumatera Barat 370.7 376.5 399.7 1.6 3.35
Riau 1919 2290.7 2430.5 19.4 20.4
Jambi 647 693 736.1 7.1 6.18
Sumatera Selatan 873.8 923 988.4 5.6 8.3
Bengkulu 308.1 293.8 298.2 -4.6 2.5
Lampung 123.4 184.9 213.6 49.8 1.79
Kep. Bangka Belitung 186.1 206.2 218 10.8 1.83
Kep. Riau 8.7 19 20.2 118.4 0.17
Dki Jakarta 0 0 0 0 0
Jawa Barat 14.1 13.6 14.3 -3.5 0.12
Jawa Tengah 0 0 0 0 0
Di Yogyakarta 0 0 0 0 0
Jawa Timur 0 0 0 0 0
Banten 14.8 19.7 21.4 33.1 0.18
Bali 0 0 0 0 0
Nusa Tenggara Barat 0 0 0 0 0
Nusa Tenggara Timur 0 0 0 0 0
Kalimantan Barat 700.5 936.4 1455.2 33.7 12.21
Kalimantan Tengah 1008.4 1115.9 1183.7 10.7 9.93
Kalimantan Selatan 424.8 512.9 437.6 20.7 3.67
Kalimantan Timur 657.3 733.4 933.9 11.6 7.84
Kalimantan Utara 0 153.3 168.7 0 1.42
Sulawesi Utara 0 0 0 0 0
Sulawesi Tengah 93.8 147.9 157.8 57.7 1.32
Sulawesi Selatan 27.9 50.9 56.4 82.4 0.47
Sulawesi Tenggara 44.8 45.2 49.4 0.9 0.41
Gorontalo 0 4.3 12.3 0 0.1
Sulawesi Barat 95.2 106.4 111.8 11.8 0.94
Maluku 0 10.3 10.6 0 0.09
Maluku Utara 0 0 0 0 0
Papua Barat 20.1 49.6 55.5 146.8 0.47
Papua 39.5 51.4 54.2 30.1 0.45
Indonesia 9102.3 10754.8 11914.5 18.2 100
Source: Badan Pusat Statistik.
Back to Section 2.
39
Table A2: Relationship between district-level oil palm production and satellite oil palm area
Sumatra East
Coefficient Std. err. Coefficient Std. err.
Palm area in 2000 4.00 0.40 4.91 0.87
Palm area in 2005 3.53 0.32 3.19 0.31
Palm area in 2010 2.81 0.29 1.55 0.17
Palm area in 2015 2.52 0.40 1.00 0.17
Source: Authors’ calculation from INDO-DAPOER and Pirker et al. (2016).
The table shows coefficients and standard errors from regressing district-level oil palm production in 2010 on area
under oil palm production based on satellite data for the years indicated in the rows. Separate regressions were run for
districts in Sumatra and those in Eastern Indonesia (“East”). All coefficients are statistically significant at 5% level.
(Back to Section 3.)
Table A3: Relationship between palm area in 2015 and distance to nearest oil palm mill and
concession
(1) (2)
Dist to mills Dist to conc.
Nearest distance (in logs) -0.105*** -0.0908***
(0.00778) (0.00907)
cons 0.423*** 0.380***
(0.0261) (0.0309)
N37173 37173
Source: Authors’ calculation from Pirker et al. (2016) and Global Forest Watch (n.d.).
The table shows results from regressing village oil palm area in 2015 on distance (in logs) to nearest oil palm mill and
area under oil palm concession. All regressions include district fixed-effects. Distances area calculated as distance
between centroids. Village boundaries are based on 2014 definitions. Standard errors clustered at the village level in
parenthesis.
*<.1 ** <.05 *** <.01. (Back to Section 3.)
40
Table A4: Village issues identified in survey data
Variables Description
Theft - home Theft of ownership of villagers’ homes or of the garden around the house of the villagers
Theft - crop Theft of ownership of crops from villagers’ fields
Theft - motor Theft of a motorcycle or motor vehicle
Robbery Robbery (ambush) on the road or when transporting crops
Low-level vio-
lence
Physical violence (involving 3 people or less); physical violence involving sharp instruments (eg
knives, swords) (involving 3 people or less); or Physical violence involving blunt instruments (in-
volving 3 people or less)
Mass physical vi-
olence
mass physical violence (involving 4 or more people); Mass physical violence involving blunt instru-
ments (involving 4 or more people); mass physical violence involving sharp instruments (eg knives,
swords) (involving 4 or more people); Physical violence involving firearms (including firearms) (in-
volving 3 people or less); Mass physical violence involving firearms (including firearms) (involving
4 or more people)
Violent crime
Violence that leads to physical injury with 1 victim; Violence that leads to physical injury with more
than 1 victim; Murder with 1 victim; Murder with more than 1 victim
Other crimes Domestic violence (eg a husband hits his wife); Sexual violence or rape; Others
Source: Authors’ compilation from survey codebook.(Back to Section 3.1, Results Table 1)
41
Table A5: Definition of conflict and violence in PODES data
Variables Description
Conflict occur-
rence
Incidence of mass fights (perkelahian massal) in the village over the past year. These details are
intended to find out mass fights between residents, students, tribes, or people others in the vil-
lage/kelurahan/nagari during the past year caused by the seizure of assets, power, women, ideo-
logical/belief differences, sports, crowds/shows entertainment, and more. The fights recorded here
are the fights that took place in the village/this village/nagari, although the perpetrators, victims and
material losses were not suffered by this village.
Type of conflict
Between commu-
nity groups
Fights between groups of citizens and other groups of residents in one village/kelurahan/nagari.
Inter-village Fights between villagers/kelurahan/nagari with residents outside the village/kelurahan/nagari (other
villages/kelurahan/nagari).
With security
forces
Fights between villagers/kelurahan/nagari and security forces.
With government
officials
Fights between villagers/kelurahan/nagari with government officials.
Students Fights between students of a school and students other school students.
Types of crime
Ordinary theft Taking of goods and or money belonging to people without the knowledge and permission of the
owner against the law.
Violent theft
Act of taking goods or livestock animal does not belong to him with the intention of possessing it
against the right. These crimes include: Theft of all types of ruminant animals (buffaloes, cows,
goats), one-hoofed animals (horses, donkeys) and pigs; Theft committed at night in a house or in a
around the grounds; Theft committed by someone by dismantling, breaking, climbing, or by using a
false key, false order, false clothing or position in order to enter the victim’s residence. Also included
in this category is pickpocketing carried out by damage the victim’s bag.
Fraud/embezzlement
Persecution/violence
Act that intentionally causes damage to the health of others, starting from those that do not cause
obstacles to the victim, injuries/disability, or become sick so that they are unable to carry out daily
activities perfectly.
Burning Act of intentionally burning something, such as a house, car and ships, which may pose a danger to
the property, life or body of another person.
Rape/crimes
against decency
Forced sexual intercourse.
Drug
abuse/trafficking
Act of abusing or consuming drugs for fun.
Gambling
Murder Act of taking another person’s life, whether intentional or not accidental.
Human traffiking
Recruitment, transportation, transfer, harboring or receiving a person, by means of the threat or use of
force or other forms of coercion, kidnapping, forgery, fraud, abuse power or position of vulnerability
or giving or receiving payments or benefits thereby obtaining the consent of a person having control
over another person, to exploitation purposes. Exploitation at least includes exploitation through the
prostitution of others or other forms of sexual exploitation, forced labor or services, slavery or other
practices similar to slavery, servitude or taking organs.
Source: PODES Survey.(Back to Section 3.2)
42
Table A6: Definition of conflict in NVMS data
Type Description
Resource Conflict
Violence triggered by resource disputes (land, mining, access to employment,
salary, pollution, etc.)
Governance Conflict
Violence is triggered by government policies or programs (public services, corrup-
tion, subsidy, region splitting, etc.)
Conflict of Election and Position Violence triggered by electoral competition or bureaucratic appointments.
Conflict of Identity Violence triggered by group identity (religion, ethnicity, tribe, etc).
Popular justice
Violence perpetrated to respond to/punish actual or perceived wrong (group vio-
lence only)
Violence in Law Enforcement
Violent action taken by members of formal security forces to perform law-
enforcement functions (includes use of violence mandated by law as well as vi-
olence that exceeds mandate for example torture or extra-judicial shooting).
Criminality Criminal violence not triggered by prior dispute or directed towards specific targets.
Domestic Violence
Domestic violence comprises of acts of violence committed by a family mem-
ber against other family member(s), where the family members live under one
roof/same household.
Separatism
Violence triggered by efforts to secede from the Unitary State of the Republic of
Indonesia (NKRI).
Other conflicts Violence triggered by other issues
Source: NVMS Coding Manual.(Back to Section 3.3)
43
Table A7: Characteristics of respondents of the primary survey by oil palm production intensity
No palm Low palm Mid palm High palm Total
Age 41.78 41.21 42.73 40.74 41.49
Male 0.50 0.50 0.50 0.50 0.50
Education:
Below middle 0.32 0.40 0.51 0.37 0.38
Middle school 0.22 0.26 0.18 0.24 0.23
Senior or above 0.46 0.34 0.31 0.38 0.39
Monthly HH income level:
Up to IDR 1mn 0.38 0.27 0.30 0.18 0.29
IDR 1 to 2mn 0.35 0.40 0.46 0.37 0.38
Above IDR 2mn 0.26 0.32 0.23 0.45 0.32
Religion:
Islam 0.71 0.86 0.95 0.96 0.85
Catholic 0.21 0.09 0.03 0.01 0.10
Protenstant 0.04 0.03 0.00 0.01 0.02
Hindu 0.01 0.00 0.01 0.00 0.00
Others 0.00 0.00 0.00 0.00 0.00
Occupation:
Rice farming 0.18 0.27 0.13 0.07 0.16
Palm farming 0.04 0.12 0.08 0.22 0.11
Rubber farming 0.11 0.21 0.43 0.22 0.21
Coffee farming 0.23 0.00 0.00 0.01 0.09
N1920
Source: Authors’ calculation. IDR = Indonesian Rupiah. mn = Million.
Note: Classification of villages done by remote sensing data. No palm refers to villages in the sample without any oil
palm production. Low palm refers to villages with less than 20% oil palm coverage. Mid palm refers to villages with
20 to 40% oil palm coverage. High palm refers to villages with above 40% oil palm coverage.
(Back to section 4.1.)
44
Table A8: Impact of oil palm production on village issues (regression results)
(1) (2) (3) (4) (5)
All group Land and Labour Violence Crime Village issues
Palm intensity
Low 1.089 0.777* 1.175 0.994 0.709***
(0.157) (0.104) (0.135) (0.0875) (0.0660)
Mid 1.281* 0.956 1.182 0.816** 0.936
(0.191) (0.133) (0.149) (0.0821) (0.0929)
High 1.077 0.946 1.074 1.003 0.984
(0.150) (0.110) (0.111) (0.0819) (0.0810)
Village characteristics
Distance to mayor’s office (logs) 0.936 1.076* 0.837*** 0.979 1.069**
(0.0428) (0.0445) (0.0342) (0.0312) (0.0346)
Infrastr: Police post within 5km 0.996 1.380*** 0.901 1.103 1.201***
(0.103) (0.127) (0.0775) (0.0724) (0.0798)
Economy: Main inc source plantation 0.814 0.701*** 0.787** 0.942 0.908
(0.103) (0.0764) (0.0806) (0.0697) (0.0679)
Presence of several ethnic groups 0.874 0.948 0.815* 0.852* 1.270**
(0.120) (0.125) (0.0915) (0.0755) (0.120)
Urban=1 0.708* 0.654** 0.796 1.185 0.774**
(0.137) (0.113) (0.114) (0.123) (0.0859)
Individual characteristics
Age 0.983 1.000 1.003 0.994 1.020
(0.0187) (0.0177) (0.0164) (0.0125) (0.0129)
Age ×Age 1.000 1.000 1.000 1.000 1.000*
(0.000217) (0.000200) (0.000186) (0.000139) (0.000140)
Male 1.090 1.076 1.180** 1.096 1.155**
(0.108) (0.0922) (0.0938) (0.0647) (0.0697)
Schooling
Middle school 0.852 0.951 0.899 0.938 1.033
(0.120) (0.113) (0.0976) (0.0763) (0.0863)
Senior or above 0.955 0.991 0.931 0.943 1.164**
(0.112) (0.105) (0.0917) (0.0693) (0.0869)
Rice 0.670** 0.502*** 0.948 0.874 0.666***
(0.122) (0.0866) (0.127) (0.0838) (0.0669)
Palm 0.840 1.239 1.112 1.027 1.053
(0.152) (0.173) (0.152) (0.106) (0.110)
Rubber 1.118 1.043 1.104 1.272*** 1.110
(0.153) (0.134) (0.132) (0.112) (0.0990)
Coffee 0.634 0.741 1.216 0.747** 0.737**
(0.180) (0.140) (0.206) (0.0960) (0.0940)
Observations 1917 1917 1917 1917 1917
Exponentiated coefficients; Robust standard errors in parentheses
* p<.1, ** p<.05, *** p<.01
Source: Authors’ calculation.
Note: Columns indicate different dependent variables: “All group” combines the categories of religious conflict, ethnic
conflict, and other conflict between groups; “crime” includes juvenile delinquency, theft, thuggery, and other crimes;
“Village issues” includes problems with village leaders, corruption, and infrastructure. Full list of village issues in-
cluded in the questionnaire can be found in Table A4.
Back to Results Table 1.
45
Table A9: Crime in the village past year (full regression results)
(1) (2) (3) (4) (5)
Low violence Mass violence Violent crime Property crime Other crime
Palm intensity
Low 0.713 1.469 0.872 1.039 0.692***
(0.160) (0.375) (0.182) (0.113) (0.0606)
Mid 1.446** 2.073*** 1.263 0.597*** 0.840*
(0.266) (0.568) (0.224) (0.0890) (0.0835)
High 0.665* 1.397 0.626** 0.916 0.853*
(0.140) (0.375) (0.136) (0.0924) (0.0700)
Village characteristics
Distance to mayor’s office (logs) 0.841** 1.117 1.121 1.068 1.085**
(0.0581) (0.142) (0.0825) (0.0436) (0.0345)
Infrastr: Police post within 5km 0.935 1.357** 1.141 0.883 1.164**
(0.132) (0.201) (0.160) (0.0782) (0.0760)
Economy: Main inc source plantation 1.221 1.496** 1.129 0.864 0.892
(0.205) (0.292) (0.165) (0.0781) (0.0649)
Presence of several ethnic groups 1.068 0.944 1.041 0.899 1.114
(0.214) (0.205) (0.246) (0.0961) (0.0984)
Urban=1 1.013 1.276 1.185 0.980 1.035
(0.234) (0.322) (0.275) (0.138) (0.107)
Individual characteristics
Age 1.028 1.025 0.956 1.032* 1.009
(0.0304) (0.0500) (0.0276) (0.0169) (0.0122)
Age ×Age 1.000 1.000 1.000 1.000** 1.000
(0.000345) (0.000557) (0.000333) (0.000181) (0.000132)
Male 1.122 1.098 1.389** 0.977 1.028
(0.144) (0.184) (0.213) (0.0745) (0.0599)
Schooling
Middle school 1.301 0.867 0.852 1.054 1.012
(0.226) (0.210) (0.191) (0.106) (0.0806)
Senior or above 1.142 0.936 0.913 1.001 1.080
(0.186) (0.196) (0.150) (0.0953) (0.0786)
Rice 1.522** 1.588* 1.009 0.704***
(0.307) (0.420) (0.118) (0.0664)
Palm 1.067 1.056 1.275 0.945 0.958
(0.273) (0.285) (0.276) (0.136) (0.0978)
Rubber 1.331 1.082 0.959 1.248* 1.157*
(0.285) (0.260) (0.198) (0.144) (0.102)
Coffee 1.349 1.457 0.788 0.964 0.662***
(0.358) (0.532) (0.253) (0.159) (0.0813)
Observations 1917 1917 1619 1917 1917
Exponentiated coefficients; Robust standard errors in parentheses
* p<.1, ** p<.05, *** p<.01
Source: Authors’ calculation.
Note: Dependent variables are indicated in column heading: ‘Low violence’ means physical violence involving 3 people or less; ‘Mass violence’
means mass physical violence involving more than 3 people; ‘Violent crime’ means violence that leads to physical injury and murder; ‘Property
crime’ includes theft or robbery; ‘Others’ includes domestic violence, sexual violence, and others. This table reports marginal effects derived from
a Probit model; full regression results can be found in Appendix Table A8.
Back to Results Table 2
46
Table A10: Summary statistics of village characteristics by palm production intensity
No palm Low palm Mid palm High palm Total
Conflict
Any mass mights 0.02 0.03 0.04 0.03 0.02
Number of fights 0.03 0.04 0.05 0.04 0.03
Intra-village fights 0.01 0.01 0.02 0.01 0.01
Inter-village fights 0.01 0.01 0.01 0.01 0.01
Crime
Theft 0.35 0.45 0.47 0.49 0.37
Robbery 0.04 0.04 0.05 0.05 0.04
Fraud 0.07 0.08 0.10 0.08 0.07
Persecution 0.04 0.05 0.05 0.04 0.04
Burning 0.01 0.02 0.02 0.01 0.01
Rape 0.02 0.03 0.02 0.03 0.02
Drugs 0.09 0.13 0.13 0.14 0.10
Gambling 0.13 0.20 0.21 0.21 0.14
Murder 0.02 0.03 0.03 0.02 0.02
Other characteristics
Number of families 508.74 604.98 678.51 626.05 530.07
Distance (km) to district mayor’s office 53.62 69.48 69.62 69.87 56.44
Infrastr: Infrastr: Police post within 5km 0.51 0.39 0.38 0.40 0.49
Economy: Main inc source plantation 0.32 0.58 0.65 0.75 0.38
Presence of several ethnic groups 0.73 0.91 0.92 0.95 0.76
Observations 37209
Source: Authors’ calculation from various PODES 2014 data.
Note: Classification of villages done by remote sensing data. No palm refers to villages in the sample without any
oil palm production. Low palm refers to villages with less than 20% oil palm coverage. Mid palm refers to villages
with 20 to 40% oil palm coverage. High palm refers to villages with above 40% oil palm coverage. Only villages in
provinces of Sumatra, Kalimantan, and Papua that were part of the NVMS dataset are included in the calculations.
(Back to Section 4.2.
47
Table A11: Incidence of various type of conflict by status of oil palm production
(1) (2) (3)
All sub-districts Has palm 2005 New palm after 2005
Mean SD Mean SD Mean SD
Other conflicts 0.05 0.21 0.05 0.21 0.03 0.17
Resource Conflict 0.13 0.33 0.19 0.39 0.11 0.31
Governance Conflict 0.07 0.26 0.06 0.24 0.07 0.26
Election and Position 0.06 0.25 0.04 0.21 0.06 0.25
Conflict of Identity 0.03 0.18 0.03 0.16 0.03 0.16
Popular justice 0.25 0.43 0.30 0.46 0.19 0.40
Violence in Law Enforcement 0.18 0.38 0.21 0.41 0.20 0.40
Criminality 0.59 0.49 0.70 0.46 0.56 0.50
Domestic Violence 0.21 0.41 0.27 0.44 0.20 0.40
Separatism 0.02 0.15 0.01 0.09 0.02 0.15
Any conflict 0.69 0.46 0.78 0.41 0.66 0.47
N7580 2280 1080
Source: Authors’ calculation from NVMS data. Only subdistricts in provinces of Sumatra, Kalimantan, and Papua
regions that were part of the NVMS sample are included in the calculations. Detailed definition of conflict types is
available in Appendix Table A6.
(Back to Section 4.3.)
48
Table A12: Relationship between oil palm area and suitability
(1) (2)
Palm area 2015 Palm area 2005
Level 1 suitable -0.0668*** -0.0730***
(0.0144) (0.0146)
Level 2 suitable 0.0447*** 0.0434***
(0.0122) (0.0168)
Level 3 suitable -0.00474*** -0.00480***
(0.00124) (0.00127)
Level 4 suitable 0.112*** 0.0606***
(0.0142) (0.0123)
Level 5 suitable 0.0913*** 0.0783***
(0.0200) (0.0212)
cons -0.00217*** -0.000794
(0.000763) (0.000930)
N2384 2384
Standard errors in parentheses
* p¡.1, ** p¡.05, *** p¡.01
Source: Authors’ calculation. The table show results from regressing palm area in 2015 (Column 1) and 2005 (Column
2) on proportion of subdistrict area with various levels of suitability. Level 0 suitable is omitted. Sub-districts are
defined based on their 2000 boundaries. (Back to Section 4.3.4.)
49
Table A13: Estimation results on the impact of palm oil on conflict using suitability
(1) (2) (3) (4) (5) (6)
Resource Resource Pop. Jus. Pop. Jus. Criminality Criminality
Suitable area 0.0907 -0.101 0.0589 -0.104 0.245* 0.385***
(0.385) (0.406) (0.185) (0.230) (0.148) (0.129)
Suitable area x 2006 0.0281 0.128 -0.177 0.0677 -0.186*** -0.0775
(0.261) (0.257) (0.117) (0.152) (0.0618) (0.0750)
Suitable area x 2007 0.0880 0.118 -0.243* -0.170 -0.0591 -0.0763
(0.264) (0.237) (0.134) (0.157) (0.0649) (0.0713)
Suitable area x 2008 0.321 0.248 -0.206 -0.157 -0.132 -0.155**
(0.262) (0.233) (0.145) (0.138) (0.0867) (0.0732)
Suitable area x 2009 0.491 0.621** -0.0398 -0.0258 0.0751 -0.0242
(0.321) (0.314) (0.142) (0.133) (0.0669) (0.0819)
Suitable area x 2010 0.242 0.143 0.137 0.0717 0.146** 0.0534
(0.228) (0.248) (0.125) (0.144) (0.0737) (0.0934)
Suitable area x 2011 0.556** 0.548*** 0.215 0.273* 0.0711 0.0000349
(0.231) (0.192) (0.136) (0.157) (0.0736) (0.0930)
Suitable area x 2012 0.689*** 0.573*** 0.352*** 0.380*** 0.0408 -0.0177
(0.267) (0.205) (0.129) (0.134) (0.0773) (0.0861)
Suitable area x 2013 0.298 0.0744 0.151 0.0979 0.290*** 0.147**
(0.263) (0.216) (0.130) (0.161) (0.0763) (0.0742)
Suitable area x 2014 0.534** 0.442* 0.288** 0.303* 0.164* 0.0315
(0.251) (0.258) (0.145) (0.169) (0.0938) (0.119)
Social No Yes No Yes No Yes
Security No Yes No Yes No Yes
Election No Yes No Yes No Yes
N12134 11377 16655 15619 22440 20868
Bootstrapped standard errors in parenthesis.
*<.1 **<.05 ***<.01
Source: Authors’ calculation. The table show results from negative binomial fixed-effects regression with violent
crime as dependent variable and percent subdistrict area with high suitability as regressor of interest. Column 1 is
the baseline model. Column 2 includes the following control variables: sub-district’s share of Christians, and share
of migrants. Column 3 includes controls for share of families living in villages with plantation business. Column 4
includes controls for share of families living in villages within 5km of a police station. Each controls are also interacted
with year dummies. Column 5 includes controls for share of families living in villages that voted for Golkar and PDP.
Sub-districts are defined based on their 2000 boundaries. (Back to Section 4.3.4.)
50