the long-term effects of genocide on antisocial...
TRANSCRIPT
The Long-term Effects of Genocide on Antisocial
Preferences
Lata Gangadharan
Asadul Islam
Chandarany Ouch
Liang Choon Wang
October 2018
Abstract
We conduct a field experiment to examine the long-term effects of exposure to the Cambodian
genocide (1975–1979), during childhood and adolescence, on individuals’ antisocial behaviors.
Our research strategy uses plausibly exogenous variations in the intensity of genocide across
regions and individuals’ direct or indirect exposure to violence. The results indicate that as the
intensity of exposure to the genocide increases, individuals directly exposed to violence during
childhood and adolescence become significantly more antisocial and less altruistic in the long
term.
Keywords: Civil Conflict, Khmer Rouge, Intensity of Exposure, Social Preferences, Field
Experiment
JEL Codes: C91, C93, D74, O12
We thank Richard Akresh, Damien de Walque, participants at the Australasian Development Economics
workshops and the ESA World meetings and seminar audiences at Monash University and Cambodian
Development Research Institute for their helpful comments. Ethics (IRB) clearance for the project was obtained
from Monash University. We are grateful for generous funding support from the Australian Research Council,
AusAID (DFAT) and Monash University. The usual disclaimer applies.
Affiliations: Department of Economics, Monash Business School, Monash University, Australia (Gangadharan,
Islam and Wang); Cambodia Development Research Institute, Cambodia (Ouch).
Email addresses: [email protected]; [email protected]; [email protected];
1
1 Introduction
There is a growing body of literature documenting the social costs of civil conflicts, particularly
their impacts on prosocial and risk preferences of adults shortly after the end of the conflict.
However, less attention has been paid to the impacts of conflict on antisocial behavior.1
Antisocial behaviors, such as violating social rules, being deceitful, thieving and being reckless,
reflect explicit decisions to harm others, sometimes even when such actions create no obvious
private or societal gain. Being antisocial is therefore a distinct step away from being less
prosocial as these undeniably harmful decisions can impact the moral fiber of society, leading
to inefficiencies in social and economic exchange (Basurto et al., 2016).
Early childhood conditions have been shown to significantly affect health and economic
outcomes later in life (see, for example, Heckman et al., 2012 for a review).While the papers in
this literature have provided very useful insights, these studies do not examine whether early
childhood experiences shape social preferences during adulthood. This paper investigates the
long-term impacts of exposure to the Cambodian genocide during childhood and early
adolescence on antisocial behaviors in adulthood. Under the Khmer Rouge (KR) regime (1975–
1979), Cambodia experienced one of the worst genocide events in human history, resulting in
the deaths of about 1.7 million people (about 21% of the country’s population) and leaving
millions more traumatized. We use experimental methods in the field to assess the impact of
direct and intense exposure to civil conflict during childhood and early adolescence on
individuals’ antisocial behaviors about four decades after the genocide. We also assess the
effects of this exposure on prosocial and risk preferences since individuals may simultaneously
exhibit antisocial and prosocial traits (Basurto et al., 2016).2
The Cambodian genocide provides a unique setting to examine the impacts of direct and
intense exposure to civil conflict, during childhood and early adolescence, on antisocial
behaviors in adulthood for a number of reasons. First, death toll data collected by the
1 Blattman and Annan (2010) and Cecchi, Leuveld, and Voors (2016) provided survey-based evidence of
antagonism among those exposed to severe war violence. Blattman and Annan (2010) find that child soldiers in
Uganda display more hostile attitudes. Cecchi et al. (2016) find that among young street football players in Sierra
Leone, those that have had more intense exposure to violence were more likely to receive a yellow or red foul
card. 2 Some studies show that individuals more affected by violent conflicts tend to exhibit prosocial behaviors, such
as trust, altruism (Gilligan et al., 2014; Gneezy & Fessler, 2012; Voors et al., 2012; Whitt & Wilson, 2007) and
egalitarianism (Bauer et al., 2014), and are more socially and politically engaged (Bellows & Miguel, 2009;
Blattman, 2009; Gilligan et al., 2014). Cassar et al. (2013), Rohner et al. (2013) and Nunn and Wantchekon (2011),
on the other hand, highlight the negative consequences of exposure to conflict on trust, fairness and willingness to
engage in impersonal exchanges. Risk preferences among individuals affected by war or political violence have
also been examined (Callen et al., 2014; Jakiela & Ozier, 2015; Voors et al., 2012). We discuss this literature in
more detail in Section 3.
2
Cambodian Genocide Program provide an independently verified measure of intensity of
exposure to the genocide. Second, given the scale and extent of the genocide, there is
considerable variation in the intensity of genocide across geographical areas, as well as in
individuals who directly witnessed violence. Third, because our research subjects were exposed
to the genocide as children, 34 to 39 years ago, our empirical strategy allows us to estimate the
long-term effects depending on individuals’ personal exposure to the genocide and the
concentration of violence in their geographic locations. In our estimation strategy, we exploit
geographical variations in intensity of exposure to the Cambodian genocide and individuals’
direct and indirect experiences of violence during this period to identify the effects of the
conflict.
The participants in our experiments consist of individuals who were both directly and
indirectly exposed to the genocide. The directly exposed individuals include people born before
or during the KR regime (1960–1979 birth cohorts) and those who reported experiencing or
witnessing violence during this period. The indirectly exposed group consists of individuals
who did not directly witness or experience this violence and includes both individuals who were
born before or during the KR regime and individuals who were born after (1979–1982 birth
cohorts).3 These participants take part in a number of behavioral games, including money-
burning (capturing financially vindictive behavior), self-reporting (dishonesty), dictator (with
taking and giving), trust, and risk games which enable us to measure their antisocial behaviors
as well as other preferences such as prosocial behaviors, and risk attitudes. To identify
individuals who were directly exposed to violence, we conduct surveys to collect data on
personal experiences relating to violence during the KR period. Further, to examine the
sensitivity of our results, we consider different birth cohorts in the directly and indirectly
exposed groups.
We find that exposure to genocide during childhood and adolescence has different long-
term effects on antisocial behaviors depending on the intensity of the genocide and whether an
individual is directly or indirectly exposed to violence. In particular, our results show that direct
exposure to violence in areas with more intensive killings led to more financially vindictive,
more opportunistic behaviors, whereas indirect exposure to violence in areas with more
intensive killings made individuals less dishonest. We find that individuals’ prosocial behaviors
3 Note, however, while these individuals did not directly witness or experience violence, they were either present
during the genocide period or experienced its aftermath. Our definition of levels of exposure builds on that
proposed by Bauer et al. (2014) for Sierra Leone. We use the Cambodian Genocide Database (CGD) on deaths of
individuals in different regions to go a step further and exploit the geographical intensity of war. In sections 4.3
and 4.4, we explain in detail our definitions of direct and indirect exposure and intensity of exposure to genocide.
3
and risk attitudes are also affected by direct and intense exposure to violence and they become
less altruistic and risk-seeking.
Our main findings are robust across a variety of sensitivity checks and specifications.
In particular, we demonstrate that our results are not affected by any differences in the observed
characteristics of participants, such as age or education. Our results are also robust to alternative
assumptions about the reliability of early childhood memory, potential survivor bias, the
inclusion (or not) of post-genocide birth cohorts and the inclusion (or not) of controls for post-
genocide relocation.
This study provides unique insights into the lasting effects of conflict and violence on
social behaviors and risk. It builds on recent evidence related to the impacts of civil conflicts
on the social behaviors of affected individuals and contributes to the literature in a number of
ways. Unlike most of the prior studies in this field, our study focuses on the long-term effects
of different degrees of exposure to genocide during childhood and early adolescence on
antisocial behaviors in adulthood. Existing literature tends to focus on the impact of recent civil
conflicts and on the link between civil conflicts and prosocial behaviors and risk preferences.
A number of experimental studies show that individuals’ social preferences develop over the
course of childhood and adolescence (Eckel et al., 2011; Fehr et al., 2013; Harbaugh et al.,
2002; Sutter & Kocher, 2007; Sutter et al., 2013). Exposure to violent conflicts during this
crucial stage, during which significant behavioral development takes place and after which
behavior becomes less malleable, can have long-term repercussions.
Social psychologists have documented the impacts of exposure to violence during
childhood and adolescence on several social and behavioral outcomes, such as delinquency and
aggression (Dubow et al., 2009; Loeber, 1990; Miller et al., 1999; Slone et al., 1999). However,
there is little evidence of the impacts of conflict and exposure to violence during childhood and
early adolescence on dishonesty and financially vindictive behaviors decades later. Dishonesty
encourages corruption (Collier et al., 2003) and creates fraud and inefficiencies that are harmful
to development. Similarly, vindictive behaviors in the form of financial sabotage are costly to
the economy (Murphy, 1993). This study provides new evidence, using incentivized economic
experiments, that exposure to violence during childhood and adolescence can have lasting
impacts on the prevalence of these destructive behaviors depending on how individuals are
exposed. The antisocial actions in our experiment include, relevant yet underexplored,
situations in which individuals are willing to incur a personal cost to harm others, with no
financial benefit to themselves or to society.
4
Our findings contribute to current knowledge by showing that direct and greater
exposure to genocide during childhood and early adolescence makes individuals significantly
more vindictive and opportunistic later in life. Indirect exposure, on the other hand, leads to
less dishonesty. Specifically, as the intensity of exposure increases, the difference between the
directly and the indirectly exposed tends to increase, such that individuals who are not directly
exposed are more likely to be less dishonest.
Finally, by distinguishing the degree of individual exposure (direct and indirect) from
the degree of the intensity of genocide in one’s region, we differentiate between different types
of conflict experiences of children and how these experiences impact long-term behavior. Thus,
our estimates potentially inform the differential effects between large-scale conflicts where
children are directly and continuously exposed to violence and killings and small-scale isolated
conflicts where children are indirectly exposed.
The paper proceeds in the following way. In Section 2, we provide a brief background
of the genocide in Cambodia. In Section 3, we present a conceptual framework that relates
conflict with social preferences and risk to the direct exposure and indirect exposure channels
and informs our research design and hypotheses. Section 4 discusses the research strategy.
Sections 5 and 6 report the results, and Section 7 concludes.
2 Background of the Cambodian Genocide
The KR took power in 1975 and ruled Cambodia until 1979. Between April 1975 and January
1979, Cambodia suffered a period of genocide under the KR regime, which was characterized
by massive destruction, violence, torture and death. The KR attempted to impose an extreme
form of Maoist communism, forced all citizens to participate in rural work projects, often
without adequate food, to build a new Cambodia based on an agrarian model. To achieve this
goal, the KR aimed to destroy traditional Cambodian social norms, cultures, religions,
organizations, networks and even family structures (Collier et al., 2003). The regime closed
schools, hospitals, and factories, abolished banking, finance and currency, isolated the country
from all foreign influences and barred Western medicine (UNESCO, 2011). It banned all
religions, and people seen taking part in religious rituals or services were executed. The KR
also confiscated all private property and relocated people from urban areas to collective farms
through agricultural labor.
The regime forced people to live in communal work camps under the control of the
Angka (organization) with extremely strict rules and policies and no freedom of movement.
Under the KR regime, most Cambodians had to work long hours each day with insufficient
5
food, no material rewards, limited access to their spouses and children and very little free time
(Chandler, 2008). At the age of eight, most children were sent to live with other children,
supervised by two or three senior KR officials. Community and family members were
encouraged to spy and report on each other, which destroyed trust and established deep-rooted
fear (Collier et al., 2003). During its four years in power, the KR executed select groups it
believed to be enemies of the state. KR cadres carried out arbitrary executions and torture
against perceived subversive elements. The KR also targeted and killed suspected political
opponents, educated individuals, individuals from high social and professional classes and
individuals who did not share the KR vision for a new Cambodia. To avoid being targeted,
people hid their identities and tried to be as inconspicuous as possible.
According to the Cambodian Genocide Database established by Yale University,
approximately 1.7 million Cambodians were killed or died from starvation or exhaustion during
the KR regime. The intensity of the killing and death differed across regions of Cambodia
(Islam et al., 2017). Adult males were the demographic group most likely to die (de Walque,
2006). Many Cambodians who survived this period either were direct victims of the regime or
witnessed violence during the KR’s rule. They experienced threats to and the loss of loved ones
and faced prolonged parental absence.
3 Violence Exposure, Social Preferences, and Risk
In this section, we conceptualize how direct and indirect exposure to violence of different levels
of intensity may influence social behaviors of individuals by drawing on theories and evidence
from studies across different disciplines. The discussion informs the research design and
identifies outcome measures of interest.
3.1 Conceptual Framework
The effect of violence exposure on a person’s social preferences can be conceptualized by the
following simple production function approach:
𝐵 = 𝐹(𝑥, 𝑛),
where social behavior, B, is expressed as a function of the violence experience, x, the person
directly suffers or witnesses, and the environmental or indirect influence of violence, n, the
person experiences. The indirect influence of violence comes from the person’s parents, peers,
community, and other environmental factors. Social behavior, B, is viewed as multi-
dimensional and may be influenced in different ways given Basurto et al.’s (2016) findings that
antisocial and prosocial behaviors can coexist.
6
We are interested in understanding how the intensity of violence, v, which varies
continuously, influences social behavior through the direct experience channel and the indirect
experience channel. By differentiating social behavior with respect to v, we obtain the following
expression:
𝑑𝐵
𝑑𝑣=
𝜕𝐵
𝜕𝑥
𝜕𝑥
𝜕𝑣+
𝜕𝐵
𝜕𝑛
𝜕𝑛
𝜕𝑣
The above expression indicates that the effect of an increase in violence intensity on social
behavior manifests through the direct exposure channel,𝜕𝐵
𝜕𝑥
𝜕𝑥
𝜕𝑣, and the indirect exposure
channel, 𝜕𝐵
𝜕𝑛
𝜕𝑛
𝜕𝑣. This framework shows that for an individual who witnesses and experiences
violence (i.e., a directly-exposed person), the total effect on social behavior due to an increase
in violence intensity consists of both the direct and indirect exposure channels, while for an
individual who does not personally witness or experience violence, the total effect on social
behavior due to an increase in violence intensity consists of the indirect exposure channel only.
3.2 Negative Impact of Violence Exposure on Social Behavior
Studies in the psychological and epidemiological literature indicate that direct exposure to
violence or stress and trauma due to fear of loss, torture or violence during childhood and
adolescence can influence individuals’ behaviors and development either by provoking
aggression and negative social behaviors or by subduing positive social behaviors. For example,
Bandura’s (1973) social learning theory suggests that, during wartime, children may learn
aggressive behaviors by observing the violence committed by others, who are often aggressive
“role models” perceived as national heroes (Kerestes, 2006). Thus, according to these theories,
as intensity of violence increases, children who directly experience violence would become
more antisocial (i.e., 𝜕𝐵
𝜕𝑥
𝜕𝑥
𝜕𝑣> 0 where B measures antisocial behavior) and less prosocial (i.e.,
𝜕𝐵
𝜕𝑥
𝜕𝑥
𝜕𝑣< 0 where B measures prosocial behavior).
Children’s social behaviors may also be indirectly affected by violence without
witnessing it or experiencing it first-hand. For example, Huesmann’s (1988) social information-
processing theory suggests that children may become more aggressive during wartime given
the abundance of stimuli perceived to be threatening, leading to aggressive behavioral responses
(Kerestes, 2006). Dubow et al. (2009) argue that the constant threat of losing loved ones or
being killed during wartime can have contemporaneous and long-term impacts on affected
children’s mental wellbeing and aggressive behaviors. Similarly, Prediger et al. (2014) show
that persistent resource scarcity increases antisocial behavior, implying that conflict-induced
7
resource scarcity can also indirectly lead to antisocial behavior. Moreover, parents may also
indirectly transfer the direct negative effect of violence they face, to their children through
intergenerational transmission (Islam et al., 2017). Therefore, even though children may not
directly witness any violence, they may become more antisocial or less prosocial as a result of
these negative environmental influences. That is, 𝜕𝐵
𝜕𝑛
𝜕𝑛
𝜕𝑣> 0 when B measures antisocial
behavior and 𝜕𝐵
𝜕𝑛
𝜕𝑛
𝜕𝑣< 0 when B measures prosocial behavior.
3.3 Mixed Evidence of Violence Exposure on Social Behaviors
Using survey data from classroom assessments, Kerestes (2006) finds that children’s direct
exposure to violence in Croatia increased their aggressive behavior and decreased their
prosocial (altruistic) behavior. Using clinical case studies, McAuley and Troy (1983) report that
children affected by riots in urban areas of Northern Ireland experienced higher rates of
antisocial behavior. Farver and Frosch (1996) find that children exposed to the Los Angeles
riots of 1992 included more aggressive words and thematic content in stories than similar
children from other parts of the country who had no direct exposure to the riots. Evidence to
the contrary also exists. For example, Raboteg-Saric et al. (1994) report that after the war in
Croatia, children showed increased prosocial tendencies in classroom surveys that elicited
instances of sharing, comforting and helping each other and exhibited no change in instances
of aggressive verbal or physical behaviors. The differences in these studies findings’ could be
due to differences in the intensity of the children’s exposure to violence and whether the
exposure was direct or indirect, aspects that we focus on in our research. Specifically, it is
possible that as the intensity of violence increases, its positive effect on behavior via the indirect
exposure channel more than compensates its negative effect via the direct exposure channel.
Several studies, primarily in the economics and political science literature, have
provided interesting insights on the impact of violence and conflict exposure on prosocial
preferences. Some indicate that experiences of war and threats may lead to more cooperative,
prosocial behavior as a result of increased empathy for victims. Sympathy may also evolve
from a sense of connectedness with others. Groups with a shared fate may experience mutual
identification, which can become a source of support. Gneezy and Fessler (2012) argue that
violent intergroup conflict elicits behaviors that enhance intra-group cooperation and strengthen
the group’s chances of victory. Bauer et al. (2016) conducted a meta-analysis of nearly 20
studies from more than 40 countries to re-analyze and weigh different theoretical explanations
(economic, social and psychological mechanisms), finding a strong, persistent pattern of a
8
positive legacy of war in terms of local cooperation and civic engagement. This pattern is
observed primarily among adults—and, in some circumstances, children—who were exposed
to violence.4 Individuals affected by conflict are observed to be more politically and socially
engaged (Gilligan, Pasquale, & Samii, 2014; Bateson, 2012; Blattman, 2009; Bellows &
Miguel, 2009).5 Voors et al. (2012) find that adults who were directly exposed to violence or
who lived in communities that were violently attacked during the civil war in Burundi display
more altruistic giving to members of their communities.
Cassar et al. (2013), on the other hand, show that a greater intensity of direct exposure
to violence during the civil war in Tajikistan lowers trust within communities and decreases
willingness to engage in impersonal exchange. They argue that exposure to conflicts and
violence may reduce prosocial behaviors towards out-group in inter-ethnic or inter-group
conflicts or even towards members of own communities when group allegiances are not easily
observable, a feature of the civil war in Tajikistan.
3.4 Mixed Evidence of Violence Exposure on Risk Attitudes
Numerous studies also demonstrate that exposure to conflict and violence influence risk-taking
behaviors, since risk attitudes are determined by both individual (e.g., personality, cognitive
ability, genetics and emotions) and environmental (e.g., friends, parents and learning
circumstances) factors (Kim & Lee, 2014). A number of recent studies show that adults’ risk
preferences are altered by traumatic experiences, which can trigger fear and/or anger; however,
evidence showing the direction of this impact is mixed. Callen et al. (2014) argue that fearful
recollections induce more pessimistic likelihood judgements and find evidence consistent with
this in Afghanistan. Another study finds an association between self-reported fear and less risky
decisions (Lerner & Keltner, 2001). Jakiela and Ozier (2015) show, using evidence from
hypothetical lottery choices, that experiencing Kenya’s post-election violence sharply increased
risk aversion. Sacco et al. (2003) find that after the 9/11 terrorist attack, individuals (in a
different country) who were not directly exposed to the attacks made more conservative and
4 Bauer et al. (2014) conducted experiments with children in Georgia six months after the end of the war between
Russia and Georgia over South Ossetia. They ran the same games among adults in Sierra Leone following an 11-
year civil war that ended in 2002. They find that experiencing war during childhood and early adulthood increases
people’s motivations towards equality for or cooperation towards their own groups but less cooperation with or
antagonism against members of out-groups. 5 A number of studies in other related contexts, such as post-election violence (Becchetti, Conzo, & Romeo, 2014)
and earthquake and tsunami damage (Caló-Blanco, et al., 2015; Cassar, Healy, & Von Kessler, 2011; Rao et al.,
2011), also find that survival threats tend to enhance local cooperation.
9
less risky decisions. Voors et al. (2012), on the other hand, find that exposure to violence during
the civil war in Burundi increased risk-seeking behaviors.
3.5 The Impacts on Social and Risk-taking Behaviors Decades Later
Though adverse effects on antisocial behaviors as a result of exposure to the Cambodian
genocide during childhood and early adolescence are expected based on other findings in the
psychology literature discussed above, it is yet to be established whether we will observe effects
on opportunistic taking and destructive envious behaviors, whether these effects persist into
adulthood decades after initial exposure, and whether the effects differ by the intensity and type
of exposure. Given Basurto et al.’s (2016) findings that antisocial behaviors and prosocial
behaviors coexist and the mixed evidence in the literature, it is also unclear whether greater
antisocial tendency will necessarily mean lesser prosocial tendency. Furthermore, it is possible
that the effects due to the direct exposure channel and the indirect exposure channel as well as
the effects on antisocial and prosocial measures may differ in direction.
We hypothesize that both the severity and the proximity type (i.e., direct or indirect) of
exposure to the Cambodian genocide influence a person’s social and risk preferences, resulting
in significantly different revealed behaviors. We use the intensity of genocide at the district
level interacted with the direct exposure measure to estimate whether the extent of exposure
differentially affected the directly and indirectly exposed groups. This allows us to examine the
differential impact of exposure to the Cambodian genocide on measures of antisocial, prosocial
and risk-taking behaviors and constitutes a significant contribution to the literature.
In the realm of antisocial behaviors, we consider acquisitive aggression, destructive
envy and dishonesty, since the Cambodian genocide (like other similar civil conflicts) created
many opportunities to aggravate these behaviors. Envious attitudes are pervasive in all societies
and can impact decision-making (Beckman et al., 2002). In particular, destructive envious
behavior, which involves decreasing others’ outcomes even at a cost for oneself, can be harmful
to the operation of competitive markets and can impede innovation and economic development.
Similarly, dishonesty encourages corruption, which is well documented to be harmful to
development. To survive in an environment of civil conflict, such as during the KR, people
must often hide and lie about their families’ backgrounds and identities. Once an honest
reputation is lost, the incentive for honest behavior in the future is greatly weakened (Collier et
al., 2003). Thus, we conduct a dictator game with taking options, a money-burning game and a
self-reporting game to elicit behaviors of opportunistic taking, destructive envy and dishonesty.
The Cambodian genocide was primarily an inter-group conflict. The KR regime
10
encouraged people to spy upon and sabotage each other, potentially eroding such prosocial
preferences as trust and altruism. The consequences of being reported to the regime were often
torture and death. Survivors, hence, frequently developed tendencies towards mistrust and
selfishness. Thus, we conduct a dictator game with giving and a trust game to examine trust
and altruism within our sample. Lastly, as both fear and anger are drivers for risk evaluations
in violent settings, and both figure prominently in the Cambodian genocide, we examine
whether different levels of exposure to genocide under the KR regime are associated with
differences in risk preferences using a risk elicitation task.
4 Research Design
4.1 Recruitment of Survey and Experimental Participants
We conducted the field experiment in several locations: Phnom Penh, Cambodia’s capital city,
and six rural districts in Cambodia’s Kampong Cham province.6 The districts were selected
from a list of districts in the Cambodian Genocide Database (CGD), which includes a district
identifier for each KR mass gravesite and the estimated number of bodies in each mass grave.7
These districts were chosen to reflect the representativeness of the CGD in terms of intensity
and exposure to genocide. We show in section 4.4 that these districts are similar to the overall
CGD districts.
We recruited participants through the survey contact lists from the local statistical
organization. To encourage participation we also posted flyers at local coffee shops and in the
market place, and also advertised on social media. We hired local research assistants who
received help from school-teachers, principals and village elders to recruit participants. This
helped in ensuring the representativeness of the sample, as they were familiar with the age and
gender distribution of their communities. In all locations, participants were required to meet the
age criterion (born between 1960 and 1982). The participants in our experiment are
representative in terms of age and gender with respect to the population in Cambodia (see Table
1 and Section 4.4 for more details).
During the recruitment process, participants were informed that they would be involved
in decision making in different contexts and that they would be paid based on their decisions.
6 Kampong Cham is one of the five largest provinces in Cambodia based on population and is 123 km from Phnom
Penh. 7 The database was developed by Yale University and has been updated by the Documentation Center of Cambodia
(DC-Cam). We use both information from the original Yale database and data on additional mass gravesites and
estimates of death numbers from the DC-Cam updates. For details on the original Yale database and the
Cambodian Genocide Program, see http://www.yale.edu/cgp/ and
http://www.d.dccam.org/Database/Index1.htm for data kept by DC-Cam.
11
We invited eligible individuals to register for the experiment a week before the actual
experiment started. 659 individuals showed up on the day of the experiment. We again verified
their age, and randomly selected individuals if more than the required number of participants
(to conduct a session in a district) showed up in a particular location. Individuals who did not
participate in this experiment but showed up were given a show up fee.8 The final sample of
participants consist of 492 individuals who participated in the survey and the experiment in
February 2014.
Overall, the districts and the individual participants in the experiment are similar to the
rest of the Cambodian population of the relevant age and sex groups in terms of intensity and
exposure to genocide.
4.2 Experiment Design
Our outcome variables are drawn from incentivized experimental games. To measure antisocial
and prosocial behaviors and risk preferences, we conduct money burning, self-reporting,
dictator, trust, and risk games. The money-burning game is designed to understand participants’
inclinations to destroy, at a cost, other players’ payoffs. The self-reporting game, which uses a
simple task and requires participants to pay themselves when there is no probability of being
detected for over-payment, can be interpreted as an indicator of dishonesty. We conduct the
dictator game in two stages. The dictator game with giving is designed to measure the extent of
altruism among participants, which might indicate their concern for the well-being of others
(rather than self-interest). The dictator game with the option to give or take indicates
opportunism and selfishness. The trust game, for example, elicits the degree to which
participants can trust one another and the extent of their trustworthiness. The risk task measures
risk-taking behaviors.
These games have been used extensively in the extant literature in this field. We briefly
describe the games below. We also detail the design and procedure used in each game in
Appendix A and present the instructions seen by the participants in Appendix B. To capture
envious and financially vindictive behaviors, we use a simpler two-player version of the money-
burning game developed by Zizzo and Oswald (2001). Participants simultaneously decide how
much, if any, of the other player’s total endowment to eliminate. Participants must pay from
their own endowment to eliminate the other player’s endowment. The fee incurred for
8 The final sample of individuals who participated in each session varied slightly across locations, depending on
the geographic location of the community and number of rooms available to run a session in a school. This
difference is, however, not correlated with exposure to genocide (mortality rate at the district level).
12
eliminating the other’s endowment is charged at three levels: 5, 10 and 20% of the amount of
the other player’s endowment a player wants to eliminate.
Attitudes towards dishonesty are measured using a variant of the self-reporting matrix
task (Mazar et al., 2008). We design a simple task with pictures instead of numerical or word
tasks to accommodate Cambodia’s low literacy level. The task involves finding the picture of
a star on a sheet with 10 tables, each of which has 9 images (see Appendix B). Each participant
is given an envelope containing a sheet of 10 tables and is instructed to find the stars within one
minute and pay themselves accordingly. To ensure that there are considerable and different
opportunities for cheating, not all of the 10 tables have a star. We design two different sheets:
a sheet with 7 stars in the 10 tables and a sheet with only 4 stars in the 10 tables. These maximum
numbers are not known to the participants. The maximum number of either 4 or 7 stars per
sheet allows considerable scope for cheating, even for top performers.
In the dictator game, there are two stages: 1) giving and 2) giving or taking (List, 2007).
Each participant plays as both player 1 (dictator) and player 2 (recipient). All participants
receive an initial endowment. The dictator receives an additional endowment, while the
recipient does not. In the dictator game with giving, the dictator can transfer any positive
amount x of the additional endowment to the anonymous recipient. The recipient must simply
accept it and is only informed of how much the dictator sends if the game is selected for the
final payment. In the dictator game with giving or taking, the dictator can send the additional
endowment to other players or take other players’ initial endowments. This means that the
dictator can send either a negative amount or a positive amount. As in stage 1, the recipient is
only told the amount the dictator sends or takes if the game is selected for final payment.
We use the trust game (introduced by Berg et al., 1995) protocol in which each
participant plays as both player 1 (sender) and player 2 (receiver) and is matched with different
participants in each role. In the first stage, all participants are senders and can send any positive
amount x of the total endowment to an anonymous receiver, knowing that the experimenter
triples the amount sent, such that the receiver receives an amount of 3x. In the second stage, all
participants play as receivers. To reduce logistical issues in the field, each receiver decides on
an amount y of 3x to return to the sender for all the corresponding amounts the receiver might
receive. The sender is only informed of the amount sent back by the receiver if the game is
selected for final payment.
To elicit risk preferences, we use a risk task with a 50% chance of winning or losing
(Gneezy & Potters, 1997). Each participant receives an endowment and can invest any positive
13
amount in a risky business. The investment yields triple the amount invested with 50%
probability and nothing with 50% probability. The outcome is decided by tossing a coin.
After completing all of the experimental games, participants are requested to complete
a survey questionnaire. The survey covers information about participants’ personal
characteristics and experiences during the KR period. 9 Participants with limited reading and
writing abilities are interviewed.
They were matched anonymously with another participant for payment purposes. We
set different endowments and participation fees for Phnom Penh and for the rural areas in
Kampong Cham province. In Phnom Penh, the endowment and participation fees are set to
twice those of the rural areas, since the average earnings of workers in Phnom Penh (based on
CSES, 2011) are approximately twice those of workers in rural areas.
4.3 Measures of Genocide Exposure
We use a number of measures of genocide exposure. First, we ask participants born between
1960 and 1979 two questions in the survey: “Did you ever see or experience physical torture
during the KR regime?” and “Did you ever see someone killed during the KR regime?” If the
response to either of these two questions is ‘yes’, we classify an individual as having been
directly exposed to the Cambodian genocide. Roughly 55% of those who were born before or
during the KR regime (born between 1960 and 1979), or 40% of the overall sample, responded
positively to this question. Note that individuals who were not directly exposed to the genocide
were, by construction, indirectly exposed to it, since they were either present during the
genocide period or experienced the genocide’s aftermath. These individuals include those born
after the KR regime ended and those who responded ‘no’ to the question above.
Second, the Cambodian Genocide Database (CGD) provides information about the
intensity of the Cambodian genocide in various areas of Cambodia, which allows us to construct
a district-level measure of the intensity of genocide exposure. The construction of this measure
is detailed in Appendix C. Briefly, to construct district-level KR mortality rates, we divide the
estimated deaths under the KR in a given district (based on information from the CGD) by the
sum of the total estimated deaths under the KR, the number of living individuals (based on
Census 1998 data) residing in each district at the beginning of the KR regime (1975) and the
number of living individuals (based on Census 1998 data) born in each district during the KR
9 We also collect attitudinal questions and personality measures. These measures do not vary systematically with
the intensity of genocide exposure for either the directly exposed group or the indirectly exposed group. Our main
results from the experimental games are robust to using these measures as control variables.
14
period.10 The geographical distribution of KR mortality rates (shaded blue) is illustrated in
Figure 1. According to the information available in the CGD, KR mortality rates in the 145
districts fall between 0 and 0.857. For districts in five provinces (Kaoh Kong, Preah Vihear,
Otdar Mean Chey, Krong Kaeb and Krong Pailin; shaded white in Figure 1), no information is
available in the CGD.11
[Figure 1]
Since we ask participants whether they have lived in the same district since birth and, if
not, in which district they resided during the KR regime, we are able to link the direct exposure
measure with the intensity of exposure measure, as well as to construct an intensity of indirect
exposure measure. Thus, among those individuals who were exposed directly to violence, we
can estimate the impact using the intensity of the direct exposure. Among those individuals
who were not exposed directly to violence, the intensity of exposure measure enables us to
estimate the impact of the extent to which they were indirectly exposed to the genocide (e.g.,
through family members, neighbors and so on).
4.4 Exogenous Variations in Violence Exposure
Table 1, Panel A, reports summary statistics of participants’ demographic characteristics and t-
tests of the means between those who were directly exposed and those who were not. The right
hand side of Panel A of Table1 shows the representativeness of the sample by comparing the
demographic characteristics of the experiment sample to data from the 2011 Cambodia Socio-
economic Survey (CSES 2011). The characteristics of respondents of the two samples are
similar, as indicated by column 8 in Panel A.
[Table 1]
The mortality rate in an individual’s district of residence during the KR regime is, on
average, 22.1% for a directly exposed individual and 21.5% for an indirectly exposed individual
(p = 0.63). This result means that a person’s type of exposure (i.e., direct or indirect) to a high
level of violence is plausibly exogenous. The average ages of the directly exposed group and
the indirectly exposed group are approximately 48 and 39, respectively (p = 0.000). This
10 Since we do not have information on the number of individuals who survived the KR regime but died between
1980 and 1998 at the district level, the estimated KR mortality rates are noisy. Note that we are able to identify
districts of residence both in 1975 and at birth because Census 1998 contains information regarding birth district,
years of residence in current district and previous district of residence. Islam et al. (2017) show that alternative
ways of constructing mortality rates do not differently affect the estimated impact of KR mortality rates on
education, earnings, fertility and health measures. 11 Only three participants in our sample were in these provinces during the KR regime. We assume that the intensity
of genocide they faced during the KR regime was zero. The results are not sensitive to dropping these participants
from the sample or using the genocide intensities of the districts in which they currently live.
15
difference in age is because indirectly exposed group includes individuals who were born
before, during, or after the KR regime. On the other hand, the directly exposed group were born
before the KR regime and older individuals are more likely to directly experience and
remember violence due to being exposed to the KR regime for a longer period of time and
having more mature memories at the time of exposure.12 Individuals in the directly exposed
group are more likely to be male (55.6%) than indirectly exposed group (39.5%) (p = 0.0004).
This difference is reasonable as males were more likely to be targeted by the KR regime and
hence directly exposed to violence given past findings by de Walque (2006), Islam et al. (2017)
and Neupert and Prum (2005), who show that males were the demographic group most likely
to die under the KR regime. Note that among those who were born between 1960 and 1979 and
were alive during the KR regime, the proportion in our sample is 48.7% and similar to the
distribution in the population (47%). The average years of schooling of the directly exposed
group and the indirectly exposed group are six and eight years, respectively (p = 0.000). This
is again consistent with Islam et al.’s (2016) finding that education and age are positively
correlated because the civil conflict and genocide in Cambodia disrupted schooling.
In our regression analysis, we control for these differences and examine the robustness
of our results to excluding age and education as control variables. The percentage of married
participants is approximately 87% in the directly exposed group and 82% in the indirectly
exposed group (p = 0.144). The majority of participants in the experiment are from the Khmer
ethnic group (99% in the directly exposed group and 99.7% in the indirectly exposed group, p
= 0.342). Only a small number of Cambodian Muslims participate in the experiment.
Overall, Panel A of Table 1 suggests that the demographic and educational
characteristics of our sampled individuals are similar to those of the national representative
sample. More importantly, given that our analysis exploits the intensity of genocide measure,
the directly exposed group and the indirectly exposed group do not differ in their intensities of
genocide exposure, as shown in row 1 (KR mortality rate).
[Table 2]
Table 2 demonstrates that the variation in the intensity of the genocide across districts
is exogenous by showing that district-level KR mortality rates are uncorrelated with a range of
proxies for pre-KR social and economic conditions. In particular, Census 1962 and the
geographical information system data allow us to construct the following proxies for pre-KR
social and economic conditions: the (1) pre-KR sex ratio, (2) pre-KR population density and
12 The number of retained memories of an event increases with an individual’s age at the time of the event, and
mature memories begin forming around the age of seven (Howe, 2013).
16
(3) geographical distance of a district to an urban center.13 The estimates in panel A are based
on the full sample of districts for which the CGD provides mortality figures. The estimates in
panel B are based on the sample of districts in which the participants in this study resided during
the KR period. Both panels show that the 1962 sex ratios are unrelated to the KR mortality rates
at the district level (column 1). The KR mortality rates are also not correlated with various
measures of 1962 population density (columns 2 through 4). Similarly, district mortality rates
under the KR are not correlated with distance from the provincial capital (column 5). Panel C
shows that the variables used to test for exogeneity in panels A and B are indeed correlates of
the current economic, educational and health outcomes of older adults who were excluded from
our sample selection criteria. Thus, taken together, the evidence from panels A, B and C
suggests that geographical variations in KR mortality are exogenous.
5 Results
5.1 Descriptive Analysis
In this section, we first present the average behaviors observed in the experimental games of
the directly exposed groups and the indirectly exposed group and then separate these groups
according to the degree of intensity of the genocide during the KR regime.
In Panel B of Table 1, we report the mean outcomes and the p-values of the differences
in these outcomes between the directly exposed and indirectly exposed groups. In appendix
Figure A1, we also graphically depict the difference in outcomes. Overall, we find that the
directly exposed group exhibits greater preferences for dishonesty than the indirectly exposed
group, and lower trust, trustworthiness and altruism. We describe these results in more detail
in appendix D.
Intensity of violence exposure is a continuous variable. For simplicity, we classify
intensity of violence exposure into three categories based on the distribution of KR death rates
within our sample. The first is the low intensity category, in which KR mortality rates fall
between the 0th and 33rd percentile. The second is the medium intensity category, in which KR
mortality rates fall between the 33rd and the 67th percentile. The third is the high intensity
category areas, in which KR mortality rates are in the 67th percentile or above. This
categorization allows us to clearly demonstrate the impact of the intensity of exposure to
13 The General Population Census 1962 provides data on commune-level population by gender. First, we match
the commune codes in Census 1962 with the district codes in Census 1998. Next, we match the commune-level
population with the district-level population based on the district codes in Census 1998. To merge Census 1962
with the CGD, we replace the sex ratios and population densities for district codes not available in Census 1962
with neighboring districts’ sex ratios and population densities.
17
violence before delving into the richer and more continuous variation in the intensity measure
in a regression framework.
We construct nine outcome variables from the experimental games described above.
The first is a variable for vindictive behavior, measured by a dummy variable that takes the
value of 1 if a person burns other players’ money for at least one of the three prices of burning
(5, 10 or 20%) in the money-burning game. The second is a binary variable measuring
dishonesty, which takes the value of 1 if a participant takes more money than that to which he
or she is entitled. The third is a variable for opportunistic taking, which takes the value of 1 if
an individual takes some or all of the other player’s endowment in the dictator game. The fourth
is a variable for altruism, measured by the percentage given to other players in the dictator
game. The fifth is a variable for trust, measured by the percentage sent to other players in the
trust game. The sixth is a variable for trustworthiness, measured by the percentage returned
from other players in the trust game. The seventh is a variable for risk preferences, captured by
the percentage of endowment a person invests in the risk game.
We also construct a simple index for antisocial and prosocial behavior using the
approach outlined by Gneezy, Leibbrandt and List (2016). The index for antisocial behavior
correspondingly comprises the behaviors in the dictator game with opportunities for giving or
taking, the money-burning game and the dishonesty games, as follows:
𝐼𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙 𝑎𝑛𝑡𝑖𝑠𝑜𝑐𝑖𝑎𝑙 𝑏𝑒ℎ𝑎𝑣𝑖𝑜𝑟 𝑖𝑛𝑑𝑒𝑥
= [1
3(
𝑎𝑚𝑜𝑢𝑛𝑡 𝑡𝑜𝑜𝑘 𝑖𝑛 𝑑𝑖𝑐𝑡𝑎𝑡𝑜𝑟 𝑔𝑎𝑚𝑒 𝑔𝑖𝑣𝑖𝑛𝑔/𝑡𝑎𝑘𝑖𝑛𝑔
𝑚𝑎𝑥𝑖𝑚𝑢𝑚 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑛𝑒 𝑐𝑜𝑢𝑙𝑑 𝑡𝑎𝑘𝑒)
+ 1
3(
𝑡𝑜𝑡𝑎𝑙 𝑎𝑚𝑜𝑢𝑛𝑡𝑠 𝑏𝑢𝑟𝑛𝑡 𝑜𝑓 𝑡ℎ𝑒 𝑡ℎ𝑟𝑒𝑒 𝑝𝑟𝑖𝑐𝑒𝑠 𝑖𝑛 𝑚𝑜𝑛𝑒𝑦 𝑏𝑢𝑟𝑛𝑖𝑛𝑔 𝑔𝑎𝑚𝑒
𝑚𝑎𝑥𝑖𝑚𝑢𝑚 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑛𝑒 𝑐𝑜𝑢𝑙𝑑 𝑏𝑢𝑟𝑛)
+1
3(
𝑎𝑚𝑜𝑢𝑛𝑡 𝑡𝑜𝑜𝑘 𝑖𝑛 𝑠𝑒𝑙𝑓 − 𝑟𝑒𝑝𝑜𝑟𝑡𝑖𝑛𝑔 𝑔𝑎𝑚𝑒
𝑚𝑎𝑥𝑖𝑚𝑢𝑚 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑛𝑒 𝑐𝑜𝑢𝑙𝑑 𝑡𝑎𝑘𝑒)] × 100
The index for prosocial behavior is composed of participants’ behaviors in three
experiments (amount sent in the trust game, total amount returned in the trust game for all
possible levels of trust and amount sent in the dictator game with giving), as follows:
18
𝐼𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙 𝑝𝑟𝑜𝑠𝑜𝑐𝑖𝑎𝑙 𝑏𝑒ℎ𝑎𝑣𝑖𝑜𝑟 𝑖𝑛𝑑𝑒𝑥
= [1
3(
𝑎𝑚𝑜𝑢𝑛𝑡 𝑠𝑒𝑛𝑡 𝑖𝑛 𝑡𝑟𝑢𝑠𝑡 𝑔𝑎𝑚𝑒
𝑚𝑎𝑥𝑖𝑚𝑢𝑚 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑛𝑒 𝑐𝑜𝑢𝑙𝑑 𝑠𝑒𝑛𝑑)
+ 1
3(
𝑎𝑙𝑙 𝑝𝑜𝑠𝑠𝑖𝑏𝑙𝑒 𝑎𝑚𝑜𝑢𝑛𝑡 𝑟𝑒𝑡𝑢𝑟𝑛𝑒𝑑 𝑖𝑛 𝑡𝑟𝑢𝑠𝑡 𝑔𝑎𝑚𝑒
𝑡𝑜𝑡𝑎𝑙 𝑝𝑜𝑠𝑠𝑖𝑏𝑙𝑒 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑛𝑒 𝑐𝑜𝑢𝑙𝑑 𝑟𝑒𝑐𝑒𝑖𝑣𝑒)
+1
3(
𝑎𝑚𝑜𝑢𝑛𝑡 𝑠𝑒𝑛𝑡 𝑖𝑛 𝑑𝑖𝑐𝑡𝑎𝑡𝑜𝑟 𝑔𝑎𝑚𝑒 𝑔𝑖𝑣𝑖𝑛𝑔
𝑚𝑎𝑥𝑖𝑚𝑢𝑚 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑛𝑒 𝑐𝑜𝑢𝑙𝑑 𝑠𝑒𝑛𝑑)] × 100
Figure 2: Means of behavioral outcomes of directly exposed and indirectly exposed
groups according to degree of exposure to violence during the KR regime
Figure 2 shows that the directly exposed group tends to exhibit lower antisocial and
prosocial preferences and risk attitudes than the indirectly exposed group at low levels of
violence exposure. As the intensity of violence exposure increases, the directly exposed group
engages in more taking, more money-burning, and more risk-seeking. By contrast, the
behaviors of the indirectly exposed groups generally does not vary much with the extent of
violence exposure, with the exception of the behavior of dishonesty, which decreases with the
intensity of violence experienced by the participants’ communities and then levels off.
40
50
60
70
% B
urn
ers
Low Medium HighIntensity of KR mortality
Money Burning Game1
02
03
04
05
0
% C
hea
ters
Low Medium HighIntensity of KR mortality
Self-Reporting Game
20
30
40
50
% T
akers
Low Medium HighIntensity of KR mortality
Dictator Game: Give or Take
10
20
30
40
Ave.
%G
ive
n
Low Medium HighIntensity of KR mortality
Dictator Game: % Given
10
20
30
40
Ave.
%S
en
t
Low Medium HighIntensity of KR mortality
Trust Game: % Sent
10
20
30
40
Ave.
%R
etu
rne
d
Low Medium HighIntensity of KR mortality
Trust Game: % Returned
40
50
60
70
Ave.
% I
nveste
d
Low Medium HighIntensity of KR mortality
Risk Game: % Invested
10
12
14
16
Ave.
Inde
x
Low Medium HighIntensity of KR mortality
Anti-social Pref. Index
18
24
30
36
Ave.
Inde
x
Low Medium HighIntensity of KR mortality
Pro-social Pref. Index
Indirectly Exposed Directly Exposed
19
5.2 Regression Specification
We are particularly interested in examining the differential effects of direct and indirect
exposure to violence of different intensity levels. We exploit the exogenous variation in the
intensity of genocide across different geographic locations. Furthermore, individuals within a
given region have either direct or indirect experiences of the conflict. Hence, we use
interactions of these two variables as a source of identification. Such an interaction term gives
us individual-specific variation in exposure (directly or indirectly) as well as intensity of
exposure that map to the framework discussed in section 3. Thus, we rely on an interaction term
of (i) intensity of exposure; and (ii) whether an individual was directly exposed to the
violence.14 In Section 4.3 and Section 4.4, we have already shown that the intensity of exposure
term is plausibly exogenous and that the intensity is similar across the directly and indirectly
exposed groups. Below we examine the effects of genocide exposure, using a regression
framework to control for covariates that could potentially correlate with outcomes:
𝑌𝑖𝑘𝑗 = 𝛽0 + 𝛽1𝐾𝑅𝑘 + 𝛽2𝐸𝑥𝑝𝑖𝑘𝑗 + 𝛽3𝐸𝑥𝑝𝑖𝑘𝑗 ∗ 𝐾𝑅𝑘 + 𝛿′𝑿𝑖𝑘𝑗 + 휀𝑖𝑘𝑗 (1)
where 𝑌𝑖𝑘𝑗 includes the experimental behavioral outcomes for individual i who resided in
district k during the KR regime and participates in the experiment in district j. The dummy
variable 𝐸𝑥𝑝𝑖𝑘𝑗 takes the value of 1 if individual i was directly exposed to genocide and the
value of 0 if individual i was indirectly exposed. A set of control variables 𝑿𝑖𝑘𝑗 includes age,
gender and education for individual i, as well as an indicator for whether the experiment took
place in Phnom Penh (dummy for urban). 휀𝑖𝑘𝑗 is the error term. Standard errors are clustered at
the level of the district of residence at the time of the KR regime. As we mentioned before,
individuals who were directly exposed to the genocide are somewhat older, less educated and
more likely to be male than those who were not directly exposed to the genocide. Thus,
controlling for these variables takes these differences into account. We have also addresses
identification using a strategy that relies solely on age and cohort differences. The results can
be interpreted as the average differences in social and risk behavior between those who were
directly affected vs. not directly affected. We omitted the results in this version of the paper for
14 Our empirical strategy is similar to Akresh, Lucchetti and Thirumurthy (2012), who use intensity of exposure
and whether someone was exposed or not-exposed (or duration of exposure) to violence to identify the effects on
health outcomes.
20
brevity. While the magnitudes of the coefficients are not comparable the results are qualitatively
similar to what we report in the paper.15
Let us now examine the interpretation of the regression coefficients. (𝛽1 + 𝛽3)
corresponds to the total effect of violence exposure on the behavior of directly-exposed group
as the intensity of violence increases (as discussed in section 3.1). 𝛽1 captures the indirect
(environmental) effect of violence exposure (𝜕𝐵
𝜕𝑛
𝜕𝑛
𝜕𝑣) while 𝛽3 captures the direct effect of
violence exposure (𝜕𝐵
𝜕𝑥
𝜕𝑥
𝜕𝑣). The sum of these two effects measures how a directly-exposed
individual’s antisocial behavior, prosocial behavior, or risk-taking behavior varies with the
intensity of genocide exposure. For an indirectly-exposed individual, 𝛽1 measures the total
effect of violence exposure on behavior as the intensity of violence increases. Since indirectly-
exposed individuals did not personally witness or suffer any violence, the total effect of
violence exposure for them measures how their antisocial behaviors, prosocial behaviors, and
risk attitudes respond to environmental influences as intensity of violence exposure increases.
The coefficient 𝛽3 is also of interest, as this measures how the behaviors of directly-exposed
individuals vary relative to those of indirectly-exposed individuals as the intensity of
experienced violence increases. This informs us about the counterfactual behaviors of indirectly
exposed individuals, had they been directly exposed to the genocide within their localities.
We further assess the credibility of our identification strategy in the following ways:
First, we allow the effects of indirect exposure to differ for those born before the end of the KR
regime and those born after the end of the regime. If the behaviors of these two groups of
individuals are similar, then we have evidence that both are robust control groups to the directly
exposed and that our results are unlikely to be biased by any self-reporting error in the
measurement of direct exposure (by those born before the end of the KR regime). Second, the
direct exposure measure can be sensitive to adult memories of early childhood. To explore the
robustness of the main results, we assume memories of direct violence exposure below certain
ages to be unreliable and code these cases as indirect exposure. Third, since individuals who
relocated after the KR regime provide us with additional variation in the intensity of the
violence exposure, we include in our regressions a control variable indicating whether an
individual has been living in the same district or location since birth. Robustness in the
estimates provides evidence that the estimated behavioral differences by direct exposure and
exposure intensity are not sensitive to displacement during or after the KR regime. Fourth, we
15 These results are available upon request.
21
exclude education as a control variable to examine whether genocide exposure may influence
outcomes through education. Finally, we add risk attitudes as an additional control variable
when examining the impact on social preferences because risk preferences may be correlated
with social preferences, including, particularly, measures of dishonesty and trust. We report
these measures to examine robustness in the section on sensitivity analysis.
5.3 Regression Results
We estimate equation (1) using a Probit estimation approach for antisocial behavior and an
Ordinary Least Squares (OLS) regression for prosocial behavior and risk and present the results
in Table 3. Columns 1 and 2 report the marginal effects on the likelihood of burning other
players’ money (evaluated at the means). Columns 3 and 4 present the marginal effects on
dishonesty (evaluated at the means). Column 5 reports the marginal effects on taking behavior
in dictator game (evaluated at the means). Column 6 reports the effects on altruism. Column 7
reports the effects on an individual’s trust. Column 8 reports the effects on an individual’s
trustworthiness. Column 9 reports the effects on a person’s risk-taking attitude. Column 10
reports the effects on the antisocial behavior index. Column 11 reports the effects on the index
of prosocial behavior.
[Table 3]
Of particular interests are- (i) the coefficient for KR mortality rates, which measures
how the behaviors of indirectly exposed individuals vary with the intensity of genocide, (ii) the
sum of the coefficients for the KR mortality rates, and (iii) the interaction between these
mortality rates and direct exposure (reported in the bottom row in bold), which measures how
the behaviors of directly exposed individuals vary with the intensity of the genocide. We find
that high levels of direct exposure to violence have a statistically significant impact on
individuals’ vindictive behaviors. Column 1 shows that as the intensity of violence exposure
increases, those directly exposed to the genocide are more likely to destroy others’ wealth. The
vindictive behaviors of those indirectly exposed to the genocide, however, do not vary
significantly with the intensity of violence in their communities.16
Columns 3 and 4 report the impact of exposure to violence on attitudes towards
dishonesty. These attitudes are invariant to the intensity of genocide exposure for individuals
16 Column 2 shows that these results are not sensitive to the inclusion of a binary indicator for whether a participant
is an advantaged player (i.e., received a gift in the money-burning game). Further, the magnitude of the estimated
coefficients is smaller when using different measures of the burning decision (e.g., burn in at least two prices of
burning and burn in all three prices of burning). The main effect remains negative and significant.
22
directly exposed to the genocide (bottom rows). However, as the intensity of genocide in the
community increases, individuals who were indirectly exposed become significantly less
dishonest than those who witnessed lower intensities of genocide. When we control for
differences in opportunities to cheat (i.e., the maximum number of possible correct answers,
either 4 or 7) in the regression, the results remain similar. These findings suggest that
individuals who were not directly affected by violence but witnessed intense chaos and the
aftermath of the conflict might appreciate their relative good fortune and perhaps develop a
stronger sense of moral integrity.
Columns 5 and 6 of Table 3 indicate that directly exposed individuals are more likely
to take some or all of other players’ endowments and give less when exposed to a high level of
violence, respectively.17 On the other hand, those who are not directly exposed to violence
experience little variation in opportunistic and altruistic behaviors with the intensity of the
violence experienced by the community. Thus, greater direct exposure to violence makes a
person significantly more opportunistic and significantly less altruistic.
Columns 7 and 8 in Table 3 (the bottom row) show that individuals directly exposed to
the genocide exhibit less trust and less trustworthy behaviors according to the intensity of the
genocide they experienced, but not significantly so. Similarly, for individuals indirectly
exposed to the genocide, trust and trustworthy behaviors do not vary significantly with the
intensity of the genocide experienced by their communities (first row).
Next, we examine the effects of genocide exposure on risk preferences. Column 9 of
Table 3 (the bottom row) shows that as the intensity of exposure to violence increases,
individuals who were directly exposed to the genocide invest significantly more in the risky
option. In contrast, individuals who were not directly exposed to violence do not vary their risk-
taking attitudes with the intensity of violence in their communities. Thus, our results indicate
that high levels of direct exposure to violence make individuals more risk-seeking.
Columns 10 and 11 of Table 3 present the effects of genocide exposure on the indexes
for antisocial and prosocial behaviors, respectively. Column 10 shows that antisocial behaviors
vary significantly with intensity of genocide exposure for directly exposed individuals, but not
for indirectly exposed individuals. The results indicate that a high level of direct exposure to
violence during adolescence increases antisocial attitudes and behaviors in the long term.
17 An ordered probit or a multinomial logit/probit regression model in which the dependent variable takes the value
of 0 (takes some or all of the other player’s endowment), 1 (does not give to or take from the other player) or 2
(gives some or all of the endowment to the other player), provides results similar to those in column 4. These
unreported results are available upon request.
23
Column 11 shows that prosocial behaviors do not vary significantly with intensity of genocide
exposure for either directly exposed or indirectly exposed individuals, although the intensity of
genocide exposure somewhat reduces prosocial behaviors for both groups.
Overall, the results show that members of the directly exposed group are more
financially vindictive, less altruistic, likely to take more from others and risk seeking if they
were residing in areas of intensive killing during the genocide. There is also some evidence that
indirectly exposed individuals may be less dishonest if the level of violence in their
communities was higher.
6 Sensitivity Analysis
In this sensitivity analysis section, we demonstrate that our results are robust to alternate
specifications and explanations.
6.1 Split Sample by Post-Genocide Birth Cohorts
There are two types of indirectly exposed individuals in our sample: individuals born before
and during the KR period and individuals born after the KR period. It is conceivable that the
individuals who were physically present during the genocide but did not directly experience
any violence might develop prosocial and antisocial preferences and risk attitudes that differ
from those of individuals who only heard about the violence and saw the aftermath of the
genocide years later. For example, individuals who were physically present during the genocide
are likely to have faced a constant threat of losing loved ones or being killed or tortured,
whereas individuals who were born after the KR period are not. To ascertain whether these two
types of indirectly exposed individuals respond differently to the intensity of genocide
experienced by their communities, we estimate equation (1) for the sample of individuals born
before the fall of the KR regime and estimate a variant of equation (1) that omits the direct
exposure variable and the interaction between direct exposure and the KR mortality rate for the
sample of individuals born after the end of the genocide.
[Table 4]
Panel A in Table 4 shows that the estimated marginal effects of KR mortality on both
the indirectly exposed group and the directly exposed group are similar to those reported in
Table 3. Likewise, the estimated marginal effects of KR mortality on the dishonesty measure
of the group of indirectly exposed individuals who were born after the KR period are very
similar to those reported in Table 3 (Panel B, Table 4). Thus, the marginal effects of the
intensity of genocide on the tendency for indirectly exposed individuals to be dishonest are
24
negative and the effects are almost the same for groups of individuals who were born just before
or after the genocide. These results also indicate that the both types of indirectly exposed
individuals provide similar counterfactuals and serve as good control groups for the directly
exposed individuals. Moreover, because the behaviors of these two groups of indirectly
exposed individuals are similar, it is unlikely that any self-reporting error in the direct exposure
measure for those who were physically present during the KR period might bias our results.
6.2 Reliability of Memories of Early Childhood Experience
We collect information on direct violence exposure through surveys. While this approach offers
a unique and individual-specific measure of exposure to violence, the information collected
may be sensitive to the reliability of adult memories and recalls of early childhood experiences.
For example, the literature on childhood amnesia and autobiographical memory development
indicates that children can remember very few events before the age of two and only some
events that occurred between the ages of two and three (Howe, 2013). The number of retained
memories of events increases with an individual’s age at the time of the event, and mature
memories begin forming around age seven (Howe, 2013). Bauer et al. (2014) also find that
greater exposure to war has no measurable impact on children below the age of seven, but that
the effects are pronounced beginning at around seven years of age.
To examine the sensitivity of our results, we recode direct experience of violence as 0
for the various birth cohorts whose early childhood memories of exposure to violence are less
reliable and report the re-estimated results in Table 4. We do this systematically by
progressively recoding the direct exposure variable to 0 for: (A) individuals born between 1978
and 1979 (less than 1.5 years old during the KR period); (B) individuals born between 1977
and 1979 (less than 2.5 years old during the KR period); (C) individuals born between 1976
and 1979 (less than 3.5 years old during the KR period); and (D) individuals born between 1975
and 1979 (less than 4.5 years old during the KR period). Table 5 demonstrates that our main
findings are robust irrespective of how we treat the adult memories of early childhood
experience. 18 The results suggest that our main conclusion concerning direct exposure to
genocide is driven by the sample of children and adolescents who were born before the KR
regime began (i.e. born before 1975).
[Table 5]
18 Results are similar even when we expand the birth cohorts to those younger than 6.5 years old during the KR
period.
25
6.3 Robustness to Post-Genocide Relocation
The variation in the intensity of genocide comes primarily from the number of districts in which
we conducted the experiments (7 in total) and the number of individuals who resided in districts
during the KR period or at birth that differ from the districts in which they participate in the
experiments (61 additional districts). Roughly 68% of the full sample and 66% of those born
before the end of the KR regime indicate that they have lived in the same districts since the KR
period. It is possible that post-genocide relocation might have influenced individuals’ prosocial
and antisocial behaviors and risk attitudes. For example, relocation may disrupt social
networks. It is also possible that individuals who are more mobile differ from those who are not
in terms of their social preferences and risk attitudes.
We assess the sensitivity of our results to the effects of relocation by including a
relocation dummy that takes the value of one if a respondent’s district of residence during the
KR period or at birth (if they were born after the KR period) differs from the district in which
they currently live. The results reported in Table 6 are very similar to those reported in Tables
3. Thus, our results are not sensitive to post-genocide relocation or the inclusion of individuals
who moved away from their birth districts. Finally, we also show that the estimated coefficients
are similar for a sub-sample of individuals who lived in the same district throughout their lives.
The results are reported in Appendix Table A1.
[Table 6]
6.4 Exclusion of Education and Age as Controls and Potential Survivor Bias
So far, we have included participants’ completed years of schooling and ages as controls in our
regression analyses. However, education is a potential channel through which the effects of
civil conflict exposure can affect individual preferences. Many existing studies examining civil
conflicts in different countries (e.g., Akresh & de Walque, 2008; Chamarbagwala & Morán,
2011; Dabalen & Paul, 2012; Islam et al., 2016; Leon, 2012; Shemyakina, 2011) find a negative
relationship between exposure to conflict and educational attainment. Similarly, age and
educational attainment are highly correlated in the Cambodian context (Islam et al., 2016).
More importantly, as the KR targeted prime-age and relatively more educated individuals (de
Walque, 2005) and educational attainment and age correlated with a wide range of behaviors
and outcomes, the sensitivity of our estimates to the exclusion of educational attainment and
age can provide an assessment of whether the estimates are biased by the unobserved selection
of genocide survivors. Thus, we estimate equation (1) for all measures based on the
experimental games, first excluding only completed number of years of schooling from the
26
regression and then excluding both years of schooling and age. As reported in panels A and B
of Table 7, the signs, magnitudes and significance levels of the coefficients of interest are
similar to the main results in Table 3.19 Thus, there is no evidence that the documented effects
of exposure to violence operate through education, are confounded by cohort differences, or
are potentially biased by the unobserved traits of survivors.
[Table 7]
6.5 Effects of Risk Attitudes on Behaviors in Experiment
Risk attitudes are likely correlated with prosocial and antisocial behaviors. Since we find strong
effect on risk attitudes, we include risk attitudes as a control variable when estimating the
impact on social preferences. Compared to Table 3, which shows the main results, Table 8
shows little change in the magnitudes of the estimated coefficients and the significance levels,
suggesting that exposure to genocide directly affects the individual behaviors observed in the
experiments.20
[Table 8]
7 Conclusion
KR’s violent enforcement of social engineering policies with the goal of transforming
Cambodia into a classless agrarian utopia produced one of the worst genocides in human
history. We examine the long-term effects of this genocide on the social behaviors of
individuals. To our knowledge this is the first study that explicitly examines the exposure to
violent conflict and its effects on antisocial behavior.
We exploit the intensity of genocide across different geographic locations where
individuals either directly or indirectly experienced or witnessed conflict-related violence
during their childhood and early adolescence in order to identify the impact of genocide on
social preferences and risk. First, we use plausibly exogenous variations in the degree of
concentration of deaths during the KR regime across districts in Cambodia. Second, we use
individuals’ direct and indirect personal experiences of war during the important developmental
window of childhood and adolescence. We show that personal experiences with violence are
not correlated with economic conditions and geographical characteristics.
19 Using Census 1998 data, we also find that the intensity of genocide variable is uncorrelated with the educational
attainment of individuals born between 1960 and 1966 (who had surpassed their schooling age in the post-genocide
period). The results are available upon request. 20 Our main results are also not affected by controlling for behavior in the previous games the participants played.
As described in the experimental procedure in appendix A, the games were conducted in two different orders. Our
results remain robust controlling for the game order.
27
We find that individuals who directly experienced or witnessed the genocide in areas of
intensive killings 30-40 years ago are now more opportunistic, more vindictive, less altruistic,
and more risk-taking. Overall, these people exhibit more antisocial preferences. Individuals
who lived in the same area as the genocide but who did not have direct experience with it are
found to be less dishonest when given the option to cheat. The results suggest that genocide has
differential long-term effects on prosocial and antisocial preferences and risk-taking behaviors
depending on whether an individual directly witnessed or experienced extreme violence during
the genocide period. Our findings also suggest that direct and intense exposure to genocide
during childhood and early adolescence can alter individuals’ social preferences and that the
effects persist decades after the genocide ends. Our results are robust to a variety of checks,
such as, among other sensitivity measures, restricting the sample size to different cut-off ages
for exposure to genocide, differences in education levels, living in the same locality since birth
and individual personalities.
We argue that the KR forced people to adopt norms and institutions that created feelings
of fear and horror, provoked antisocial behaviors and discouraged prosocial motivations. While
the KR regime could be dismissed as a uniquely horrific historic event of little general relevance
to the world today, the extreme ideology and the general disregard for human life which were
the defining characteristics of the Cambodian genocide and which led to human catastrophe
can still be seen in several places around the world today.21 Similar extreme events, for example,
have been experienced in China during the Cultural Revolution, in Rwanda, in Yugoslavia and,
in recent years, in Syria.
While more research is needed to explore the generalizability of our findings to other
contexts, we expect our results to improve the understanding of the long-term effects of extreme
violence and how these effects can be influenced by the proximity and intensity of violence.
The specific motivations of perpetrators may be unique to each conflict; however, the
encouragement of animosity, repression and murder on a massive scale are all common
elements. In sum, the lessons of the Cambodian genocide may apply well beyond the country’s
borders and could have implications for current events and their effects on long-term behaviors
in a post-conflict society.
21Extreme state ideologies have also been found to impact individuals’ behaviors in other (non-violent) settings.
For example, Cameron et al. (2013) find that individuals who were single children due to the One Child Policy in
China exhibit different behavioral characteristics than individuals who had siblings.
28
References
Akresh, R., Lucchetti, L., & Thirumurthy, H. (2012). Wars and child health: Evidence from the
Eritrean–Ethiopian conflict. Journal of Development Economics, 99(2), 330–340.
Akresh, R., & de Walque, D. (2008). Armed conflict and schooling: Evidence from the 1994
Rwandan genocide, Policy Research Working Paper No. 4606. Washington, DC: The
World Bank.
Bandura, A. (1973). Aggression: A social learning analysis. Englewood Cliffs, N.J.: Prentice-
Hall.
Basurto, X., Blanco, E., Nenadovic M. & Vollan, B. (2016). Integrating simultaneous
prosocial and antisocial behavior into theories of collective action, Science Advances 2
(3), e1501220.
Bateson, R. (2012). Crime victimization and political participation. American Political Science
Review, 106(3), 570–587.
Bauer, M., Blattman, C., Chytilová, J., Henrich, J., Miguel, E., & Mitts, T. (2016). Can war
foster cooperation? Journal of Economic Perspectives, 30(3), 249–274.
Bauer, M., Cassar, A., Chytilová, J., & Henrich, J. (2014). War’s enduring effects on the
development of egalitarian motivations and in-group biases. Psychological Science,
25(1), 47–57.
Bauer, P. J., & Larkina, M. (2014). Childhood amnesia in the making: Different distributions
of autobiographical memories in children and adults. Journal of Experimental
Psychology: General, 143(2), 597–611.
Becchetti, L., Conzo, P., & Romeo, A. (2014). Violence, trust, and trustworthiness: evidence
from a Nairobi slum. Oxford Economic Papers, 66(1), 283–305.
Beckman, S. R., Formby, J. P., Smith, W. J., & Zheng, B. (2002). Envy, malice and Pareto
efficiency: An experimental examination. Social Choice and Welfare, 19(2), 349–367.
Bellows, J., & Miguel, E. (2009). War and local collective action in Sierra Leone. Journal of
Public Economics, 93(11), 1144–1157.
Berg, J., Dickhaut, J., & McCabe, K. (1995). Trust, reciprocity, and social history. Games and
Economic Behavior, 10(1), 122–142.
Blattman, C. (2009). From violence to voting: War and political participation in Uganda.
American Political Science Review, 103(2), 231–247.
Blattman, C & Annan, J. (2010). The consequences of child soldiering. Review of Economics
and Statistics, 92 (4), 882-898
Callen, M., Isaqzadeh, M., Long, J. D., & Sprenger, C. (2014). Violence and risk preference:
Experimental evidence from Afghanistan. American Economic Review, 104(1), 123–
148.
Caló-Blanco, A., Kovarik, J., Mengel, F., & Romero, J. G. (2015). Natural disasters and
indicators of social cohesion. Unpublished manuscript.
Cameron, L., Erkal, N., Gangadharan, L., & Meng, X. (2013). Little emperors: Behavioral
impacts of China’s one-child policy. Science, 339(6122), 953–957.
Cassar, A., Grosjean, P., & Whitt, S. (2013). Legacies of violence: Trust and market
development. Journal of Economic Growth, 18(3), 285–318.
Cassar, A., Healy, A., & Kessler, C. (2011). Trust, risk, and time preferences after a natural
disaster: Experimental evidence from Thailand. Unpublished manuscript.
Chamarbagwala, R., & Morán, H. (2011). The human capital consequences of civil war:
Evidence from Guatemala. Journal of Development Economics, 94(1), 41–61.
Chandler, D. P. (2008). A history of Cambodia (4th ed.). Boulder, Colo: Westview Press.
29
Cecchi, F., Leuveld, K. & Voors, M. (2016). Civil War Exposure and Competitiveness:
Experimental Evidence from the Football Field in Sierra Leone. Economic
Development and Cultural Change 64 (3): 405-435.
Collier, P., Elliott, V. L., Hegre, H., Hoeffler, A., Reynal-Querol, M., & Sambanis, N. (2003).
Breaking the conflict trap: Civil war and development policy. Washington, DC: The
World Bank.
Dabalen, A. L., & Paul, S. (2012). Estimating the causal effects of conflict on education in Côte
d’Ivoire (Policy Research Working Paper No. 6077). Washington, DC: The World Bank.
De Walque, D. (2006). The socio-demographic legacy of the Khmer Rouge period in Cambodia.
Population Studies, 60(2), 223–231.
Dubow, E. F., Huesmann, L. R., & Boxer, P. (2009). A social-cognitive-ecological framework
for understanding the impact of exposure to persistent ethnic-political violence on
children’s psychosocial adjustment. Clinical Child and Family Psychology Review,
12(2), 113–126.
Eckel, C., Grossman, P. J., Johnson, C. A., de Oliveira, A. C. M., Rojas, C., & Wilson, R.
(2011). Social norms of sharing in high school: Teen giving in the dictator game.
Journal of Economic Behavior and Organization, 80(3), 603–612.
Farver, J., & Frosch, D. (1996). L.A. stories: Aggression in preschoolers’ spontaneous
narratives after the riots of 1992. Child Development, 67(1), 19–32.
Fehr, E., Glatzle-Rutzler, D., & Sutter, M. (2013). The development of egalitarianism, altruism,
spite and parochialism in childhood and adolescence. European Economic Review, 64,
369–383.
Gilligan, M., Pasquale, B., & Samii, C. (2014). Civil war and social cohesion: Lab-in-the-field
evidence from Nepal. American Journal of Political Science, 58(3), 604–619.
Glaeser, E. L., Laibson, D. I., Scheinkman, J. A., & Soutter, C. L. (2000). Measuring trust.
Quarterly Journal of Economics, 115(3), 811–846.
Gneezy, A., & Fessler, D. M. T. (2012). Conflict, sticks and carrots: War increases prosocial
punishments and rewards. Proceedings of the Royal Society B, 279(1727), 219–223.
Gneezy, U., Leibbrandt, A., & List, J. (2016). Ode to the sea: Workplace organizations and
norms of cooperation. Economic Journal, 126(595), 1856–1883.
Gneezy, U., & Potters, J. (1997). An experiment on risk taking and evaluation periods.
Quarterly Journal of Economics, 112(2), 631–645.
Guiso, L., Sapienza, P., & Zingales, L. (2004). The role of social capital in financial
development. American Economic Review, 94(3), 526–556.
Harbaugh, W., Krause, K., & Vesterlund, L. (2002). Risk attitudes of children and adults:
Choices over small and large probability gains and losses. Experimental Economics,
5(1), 53–84.
Heckman, J., Conti, G., Hansman, C., Novak, M., Ruggiero, A., & Suomi, S. (2012). Primate
Evidence on the Late Health Effects of Early-Life Adversity. Proceedings of the
National Academy of Sciences, 109(23), 8866-8871.
Howe, M. L. (2013). Memory development: Implications for adults recalling childhood
experiences in the courtroom. Nature Reviews Neuroscience, 14(12), 869–876.
Huesmann, L. R. (1988). An information processing model for the development of aggression.
Aggressive Behavior, 14(1), 13–24.
Islam, A., Ouch, C., Smyth, R., & Wang, L. C. (2016). The long-term effects of civil conflicts
on education, earnings, and fertility: Evidence from Cambodia. Journal of Comparative
Economics, 44(3), 800–820.
30
Islam, A., Ouch, C., Smyth, R., & Wang, L. C. (2017). The intergenerational effect of
Cambodia's Genocide on Children's Education and Health. Population and
Development Review, 4(2): 331-353.
Jakiela, P., & Ozier, O. (2015). The impact of violence on individual risk preferences: Evidence
from a natural experiment, Policy Research Working Paper No. 7440, (pp. 1–37). The
World Bank.
Keresteš, G. (2006). Children’s aggressive and prosocial behavior in relation to war exposure:
Testing the role of perceived parenting and child’s gender. International Journal of
Behavioral Development, 30(3), 227–239.
Kim, Y.-I., & Lee, J. (2014). The long-run impact of a traumatic experience on risk aversion.
Journal of Economic Behavior & Organization, 108, 174–186.
Knack, S., & Keefer, P. (1997). Does social capital have an economic payoff? A cross-country
investigation. Quarterly Journal of Economics, 112(4), 1251–1288.
Leon, G. (2012). Civil conflict and human capital accumulation: The long-term effects of
political violence in Peru. Journal of Human Resources, 47(4), 991–1022.
Lerner, J. S., & Keltner, D. (2001). Fear, anger, and risk. Journal of Personality and Social
Psychology, 81(1), 146–159.
Lewis, M. (2001). Issues in the study of personality development. Psychological Inquiry, 12(2),
67–83.
List, J. a. (2007). On the interpretation of giving in dictator games. Journal of Political
Economy, 115(3), 482–493.
Loeber, R. (1990). Development and risk factors of juvenile antisocial behavior and
delinquency. Clinical Psychology Review, 10(1), 1–41.
Mazar, N., Amir, O., & Ariely, D. (2008). The dishonesty of honest people: A theory of self-
concept maintenance. Journal of Marketing Research, 45(6), 633–644.
McAuley, R., & Troy, M. (1983). The impact of urban conflict and violence on children referred
to a child psychiatry clinic. In J. Harbison (Ed.), Children of the troubles: Children in
Northern Ireland (pp. 33–43). Belfast: Stranmillis College, Learning Resources Unit.
McCrae, R. R., Costa, P. T., Ostendorf, F., Angleitner, A., Caprara, G. V., Barbaranelli, C., …
Piedmont, R. L. (1999). Age differences in personality across the adult life span:
Parallels in five cultures. Developmental Psychology, 35(2), 466–477.
Miller, L. S., Wasserman, G. A., Neugebauer, R., Gorman-Smith, D., & Kamboukos, D. (1999).
Witnessed community violence and antisocial behavior in high-risk, urban boys.
Journal of Clinical Child Psychology, 28(1), 2–11.
Murphy, Kevin R. (1993). Honesty in the workplace. Pacific Grove, Calif.: Brooks/Cole
Publishing Co.
Neupert, R., & Prum, V. (2005). Cambodia: Reconstructing the demographic stab of the past
and forecasting the demographic scar of the future. European Journal of Population,
21(2), 217–246.
Neyer, F. J., & Asendorpf, J. B. (2001). Personality-relationship transaction in young adulthood.
Journal of Personality and Social Psychology, 81(6), 1190–1204.
Nunn, N. and Wantchekon, L. (2011). The slave trade and the origins of mistrust in Africa,
American Economic Review, 101(7), 3221-3252.
Prediger, S., Vollan, B., & Herrmann, B. (2014). Resource scarcity and antisocial behavior.
Journal of Public Economics, 119, 1-9.
Raboteg‐Šaric, Z., Žužul, M., & Keresteš, G. (1994). War and children’s aggressive and
prosocial behavior. European Journal of Personality, 8(3), 201–212.
Rammstedt, B., & John, O. P. (2007). Measuring personality in one minute or less: A 10-item
short version of the Big Five Inventory in English and German. Journal of Research in
Personality, 41(1), 203–212.
31
Rao, L.-L., Han, R., Ren, X.-P., Bai, X.-W., Zheng, R., Liu, H., … Li, S. (2011). Disadvantage
and prosocial behavior: The effects of the Wenchuan earthquake. Evolution and Human
Behavior, 32(1), 63–69.
Rohner, D., Thoenig, M., & Zilibotti, F. (2013). Seeds of distrust: Conflict in Uganda. Journal
of Economic Growth, 18(3), 217–252.
Sacco, K., Galletto, V., & Blanzieri, E. (2003). How has the 9/11 terrorist attack influenced
decision making? Applied Cognitive Psychology, 17(9), 1113–1127.
Shemyakina, O. (2011). The effect of armed conflict on accumulation of schooling: Results
from Tajikistan. Journal of Development Economics, 95(2), 186–200.
Slone, M., Lobel, T., & Gilat, I. (1999). Dimensions of the political environment affecting
children’s mental health: An Israeli study. Journal of Conflict Resolution, 43(1), 78–91.
Sutter, M., & Kocher, M. G. (2007). Trust and trustworthiness across different age groups.
Games and Economic Behavior, 59(2), 364–382.
Sutter, M., Kocher, M. G., Glatzle-Ruetzler, D., & Trautmann, S. T. (2013). Impatience and
uncertainty: Experimental decisions predict adolescents’ field behavior. American
Economic Review, 103(1), 510–531.
UNESCO. (2011). Edcuation and fragility in Cambodia. Paris: UNESCO-International
Institute for Educational Planning.
Voors, M. J., Nillesen, E. E. M., Verwimp, P., Bulte, E. H., Lensink, R., & Soest, D. P. V.
(2012). Violent conflict and behavior: A field experiment in Burundi. American
Economic Review, 102(2), 941–964.
Whitt, S., & Wilson, R. K. (2007). The dictator game, fairness and ethnicity in postwar Bosnia.
American Journal of Political Science, 51(3), 655–668.
Zizzo, D. J., & Oswald, A. J. (2001). Are people willing to pay to reduce others’ incomes?
Annales d’Économie et de Statistique, (63/64), 39–65.
32
Figure 1: Distribution of KR mortality rates across districts in Cambodia
Note: Blue shaded areas are districts with information of mortality under the
Khmer Rouge regime available. White circles with a red marker denote districts
in which the participants in our experiment resided during the Cambodian
genocide period.
33
Table 1: Demographic characteristics of experiment participants
Experimental sample CSES 2011
Mean differences
between Experiment and
CSES 2011 samples
Main Variables All
Std.
Dev. Range
Direct
Exposure
Indirect
Exposure
p-value
of diff.
(t-test)
All p-value of diff. (t-test)
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A
KR mortality rate 0.217 0.153 0-0.542 0.221 0.215 0.633
Age 42.31 7.290 32-54 47.8 38.7 0.000 41.88 0.194
Male (=1) 0.46 0.499 0-1 0.556 0.395 0.000 0.47 0.689
Education (years) 7.32 4.459 0-22 6.20 8.05 0.000 7.13 0.311
Married (=1) 0.84 0.369 0-1 0.87 0.82 0.144 0.82 0.477
Khmer (=1) 0.99 0.078 0-1 0.99 0.997 0.342
Observations 492 196 296 4601
Panel B
Burn 0.467 0.499 0-1 51.0 43.9 0.123
Dishonest 0.309 0.462 0-1 41.3 24.0 0.000
Take 0.350 0.477 0-1 33.7 35.8 0.627
% Given 24.309 25.259 0-100 21.1 26.4 0.023
% Sent 30.955 27.858 0-100 27.3 33.4 0.018
% Returned 30.908 24.782 0-100 28.1 32.8 0.000
% Invested 46.443 27.727 0-100 45.2 47.3 0.402
Observations 492 196 296
Note: Age originally reported in CSES 2011 is recoded to reflect the age in February 2014. Column (6) tests the differences of means reported
in columns (4) and (5). Column (8) tests the differences of means reported in columns (1) and (7).
34
Table 2: Exogeneity of mortality rates under the Khmer Rouge regime
Sex ratio
in 1962
Density
in 1962
Density —Men in
1962
Density — Women
in 1962
Distance to capital
district (1) (2) (3) (4) (5)
Panel A: CGD sample
Test of exogeneity of mortality rates under the Khmer Rouge regime
KR mortality rates -0.022 -814.459 -417.384 -397.075 4.054
(0.017) (663.015) (335.879) (327.192) (9.648)
R-squared 0.008 0.008 0.008 0.007 0.001
Observations 145 145 145 145 145
Panel B: Experiment sample
Test of exogeneity of mortality rates under the Khmer Rouge regime
KR mortality rates -0.005 -577.343 -290.417 -286.926 16.750
(0.022) (1172.329) (592.270) (580.062) (16.325)
R-squared 0.001 0.003 0.003 0.003 0.014
Observations 68 68 68 68 68
Panel C: CSES 2004 sample of individuals born before 1950
Correlation of test of exogeneity variables and actual outcomes
Mean years of
schooling
Mean monthly
earnings
Mean monthly
household income
Illness/injury during
the past 30 days Disabled
(1) (2) (3) (4) (5)
Dependent Variable:
Sex ratio in 1962 4.469** 5.728*** 5.834*** -0.314 -0.231
(1.760) (1.791) (0.794) (0.273) (0.287)
Density in 1962 (in 10,000) 3.521*** 1.140*** 1.794*** 0.105** -0.054*
(0.245) (0.334) (0.153) (0.049) (0.028)
Density—Men in 1962 (in 10,000) 6.925*** 2.233*** 3.527*** 0.206** -0.106*
(0.507) (0.675) (0.315) (0.097) (0.056)
Density—Women in 1962 (in 10,000) 7.160*** 2.327*** 3.649*** 0.216** -0.110*
(0.476) (0.662) (0.298) (0.098) (0.058)
Distance to capital district (in 10,000) -288.508*** -113.597** -148.448*** 3.122 2.543
(42.899) (43.645) (27.979) (6.579) (6.315)
Observations 141 125 141 141 141
Note: All observations are measured at the district level. The values for the dependent variables in columns 1–4 of panels A and B are from Census 1962. In panel B,
KR mortality rates are assumed to be zero for the three districts in which KR mortality data are not available (the results are similar if the three districts are excluded).
In panel C, each estimate came from a separate regression with one explanatory variable. Panel C shows whether the variables used for the test of exogeneity are
predictive of actual outcomes. Robust standard errors are in parentheses. *** p<0.01, ** p<0.05, * p<0.10
35
Table 3: Estimates of effects of exposure to genocide on prosocial and antisocial behaviour and risk Money Burning Self-reporting Dictator Trust Risk Index
Dependent variable: Burn Burn Dishonest Dishonest Take % Given % Sent % Returned % Invested Antisocial
Preferences
Prosocial
Preferences
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
KR mortality rate (β1) -0.055 -0.057 -0.678*** -0.701*** -0.244 -5.373 -9.932 -5.274 8.264 -8.650 -6.531 (0.153) (0.152) (0.216) (0.215) (0.245) (7.933) (12.105) (12.265) (8.825) (6.992) (9.972)
Direct exposure (β2) -0.077 -0.079 0.060 0.055 -0.124* -0.752 -5.014 -3.050 -5.721 -2.070 -3.151
(0.073) (0.072) (0.056) (0.056) (0.074) (4.782) (4.094) (3.226) (3.448) (2.292) (3.213)
KR mortality rate × Direct exp. (β3) 0.563*** 0.571*** 0.596** 0.587** 0.585*** -20.634 0.706 4.959 23.501** 18.627*** -4.264
(0.162) (0.159) (0.259) (0.261) (0.210) (16.011) (11.202) (13.703) (9.256) (6.947) (11.937)
Age 0.006 0.006 -0.003 -0.003 -0.001 0.070 -0.013 -0.204 -0.104 -0.071 -0.045
(0.004) (0.004) (0.003) (0.003) (0.004) (0.298) (0.201) (0.181) (0.212) (0.079) (0.176)
Education (years) 0.018*** 0.017*** -0.017*** -0.018*** 0.003 0.536 0.601* 0.280 0.980** 0.041 0.496*
(0.005) (0.005) (0.006) (0.006) (0.005) (0.381) (0.324) (0.230) (0.399) (0.131) (0.272)
Male -0.037 -0.036 -0.092** -0.097** -0.121*** 9.398*** 8.376*** 3.043** 8.900*** -2.129 6.867***
(0.050) (0.051) (0.044) (0.041) (0.039) (2.272) (2.300) (1.501) (2.451) (1.898) (1.389)
Phnom Penh -0.148* -0.147* -0.190*** -0.192*** -0.168* 18.726*** 21.808*** 13.328*** 5.883 -6.896*** 18.347***
(0.085) (0.085) (0.073) (0.072) (0.099) (4.409) (4.132) (3.329) (4.371) (2.121) (3.380)
Advantaged player -0.016
(0.043)
Maximum number of correct answers -0.122***
(0.041)
R-squared 0.140 0.132 0.059 0.088 0.039 0.180
Observations 492 492 492 492 492 492 492 4920 492 492 492
KR mortality rate +
KR mortality rate × Direct exp. (β1+ β3) 0.508*** 0.514*** -0.082 -0.114 0.341** -26.007** -9.226 -0.315 31.765*** 9.977** -10.795
(p-values) (0.000) (0.000) (0.572) (0.453) (0.011) (0.050) (0.460) (0.977) (0.000) (0.017) (0.290)
Note: Columns 1 and 2 report the marginal effects evaluated at the mean from the Probit estimation where dependent variable equals 1 if the player decides to reduce (burn) the other player’s money for at least 1 of the 3
prices of burning and 0 if player decides not to burn any amount. Columns 3 and 4 report the marginal effects evaluated at the mean from the Probit estimation where dependent variable equals 1 if the player takes extra
money to which he or she is not entitled and 0 otherwise. The advantaged player equals 1 if the player receives a gift and 0 otherwise. The maximum number of correct answers equals 1 if the total number of correct answers
in the self-reporting game is 7 and 0 if total number of correct answers is 4. Column 5 reports the marginal effects evaluated at the mean from the Probit estimation when the dependent variable equals 1 if the player decides
to take some or all of the other player’s endowment and 0 otherwise. Column 8 includes 10 observations per person as there are 10 possible values that each sender might send. Column 10 reports the index for antisocial
behaviour which comprises of the behaviour in the dictator game -giving or taking, money burning and dishonesty games. Column 11 reports the index for prosocial behaviour, which is composed of the behaviour in three experiments (amount sent in the trust game, total amount returned in the trust game for all possible levels of trust, amount sent in the dictator game - giving). Robust standard errors clustered by districts (68 districts) are
reported in parentheses. *** significant at 1%, ** significant at 5%, * significant at 10%.
36
Table 4: Splitting the Sample by Post-Genocide Birth Cohorts
Money Burning Self-reporting Dictator Trust Risk Index
Dependent variable: Burn Burn Dishonest Dishonest Take % Given % Sent % Returned % Invested Antisocial
Preferences
Prosocial
Preferences
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
Panel A: Birth cohorts: 1960-KR Period
KR mortality rate (β1) -0.100 -0.098 -0.489** -0.564*** -0.371 -1.024 -7.562 8.573 13.953 -9.208 0.618
(0.265) (0.264) (0.229) (0.217) (0.419) (9.266) (15.769) (10.004) (13.423) (10.656) (8.552)
Direct exposure (β2) -0.101 -0.105 0.110* 0.094 -0.146* 0.026 -4.527 -0.357 -5.029 -2.189 -1.744
(0.077) (0.078) (0.063) (0.062) (0.088) (4.950) (4.897) (3.308) (4.171) (2.946) (3.342)
KR mortality rate × Direct exp. (β3) 0.623** 0.632** 0.417 0.456* 0.686* -25.160* -2.083 -8.970 18.120 18.734* -11.685
(0.250) (0.254) (0.260) (0.247) (0.380) (13.044) (15.271) (12.958) (12.927) (9.954) (11.183)
Age 0.003 0.003 -0.007 -0.006 -0.002 0.120 0.069 -0.177 -0.102 -0.111 0.024
(0.006) (0.006) (0.006) (0.006) (0.003) (0.296) (0.293) (0.153) (0.276) (0.093) (0.173)
Education (years) 0.010 0.010 -0.023*** -0.024*** 0.009* 0.574* 0.554 0.195 0.954** 0.095 0.477*
(0.007) (0.007) (0.008) (0.007) (0.005) (0.300) (0.399) (0.240) (0.423) (0.165) (0.241)
Male 0.012 0.013 -0.112** -0.119*** -0.114** 9.733*** 9.505*** 4.340** 10.256*** -1.953 7.722***
(0.070) (0.071) (0.048) (0.045) (0.051) (2.589) (2.605) (1.830) (3.573) (2.404) (1.275)
Phnom Penh -0.163 -0.161 -0.280*** -0.289*** -0.135 18.832*** 20.074*** 12.592*** 7.857 -7.164*** 17.484***
(0.114) (0.113) (0.103) (0.103) (0.099) (5.946) (5.669) (4.682) (5.360) (2.069) (4.790)
Advantaged player -0.033
(0.050)
Max. no. of correct answers -0.125***
(0.045)
R-squared 0.134 0.104 0.045 0.104 0.040 0.152
Observations 358 358 358 358 358 358 358 3580 358 358 358
KR mortality rate +
KR mortality rate × Direct exp. (β1+ β3) 0.523*** 0.534*** -0.072 -0.108 0.315** -26.184* -9.645 -0.397 32.073*** 9.526** -11.067
(p-values) (0.001) (0.000) (0.572) (0.424) (0.015) (0.057) (0.450) (0.972) (0.000) (0.018) (0.304)
Panel B: Post-Genocide Birth Cohorts
KR mortality rate -0.016 -0.006 -0.706*** -0.676*** -0.146 -10.938 -12.656 -20.844 -0.024 -8.321 -14.801
(0.205) (0.208) (0.192) (0.216) (0.146) (15.167) (14.521) (17.567) (9.999) (6.676) (14.397)
R-squared 0.148 0.197 0.093 0.052 0.041 0.235
Observations 134 134 134 134 134 134 134 1340 134 134 134
Notes: See notes in Table 3 for details about the dependent variables and specifications used. Post-genocide birth cohorts are those born in 1980 and after. Robust standard errors clustered by districts (68 districts) are
reported in parentheses. *** significant at 1%, ** significant at 5%, * significant at 10%.
37
Table 5: Robustness to Potential Unreliability of Adult Memories of Childhood Experience
Money Burning Self-Reporting Dictator Trust Risk Index
Burn Dishonesty Take % Given % Sent % Returned % Invested Antisocial Prosocial
(1) (2) (3) (4) (5) (6) (7) (8) (9)
A. Birth cohorts 1978-1979 recoded
KR mortality rate -0.055 -0.678*** -0.244 -5.373 -9.932 -5.274 8.264 -8.650 -6.531
(0.718) (0.002) (0.319) (0.501) (0.415) (0.669) (0.352) (0.220) (0.515) KR mortality rate +
KR mortality rate × Direct exp. 0.508*** -0.082 0.341** -26.007** -9.226 -0.315 31.765*** 9.977** -10.795
(0.000) (0.572) (0.011) (0.050) (0.460) (0.977) (0.000) (0.017) (0.290) B. Birth cohorts 1977-1979 recoded
KR mortality rate -0.049 -0.666*** -0.237 -5.793 -10.099 -5.570 7.922 -8.410 -6.823
(0.749) (0.002) (0.336) (0.466) (0.405) (0.650) (0.378) (0.238) (0.494) KR mortality rate +
KR mortality rate × Direct exp. 0.499*** -0.099 0.330** -25.316* -8.878 0.205 32.397*** 9.620** -10.277
(0.001) (0.482) (0.010) (0.053) (0.481) (0.985) (0.000) (0.017) (0.312) C. Birth cohorts 1976-1979 recoded KR mortality rate -0.042 -0.617*** -0.255 -5.490 -8.489 -5.603 7.016 -8.645 -6.210
(0.772) (0.003) (0.291) (0.496) (0.521) (0.649) (0.435) (0.223) (0.553) KR mortality rate +
KR mortality rate × Direct exp. 0.495*** -0.136 0.361*** -26.115* -11.308 0.228 33.91*** 10.178** -11.319
(0.001) (0.269) (0.008) (0.052) (0.321) (0.983) (0.000) (0.016) (0.261) D. Birth cohorts 1975-1979 recoded KR mortality rate -0.020 -0.631*** -0.214 -5.970 -9.611 -6.697 6.449 -7.981 -7.082
(0.892) (0.001) (0.354) (0.468) (0.468) (0.593) (0.468) (0.253) (0.507)
KR mortality rate + KR mortality rate × Direct exp. 0.475*** -0.107 0.334** -26.339* -9.845 1.937 35.369*** 9.900** -10.360
(0.001) (0.364) (0.040) (0.067) (0.396) (0.860) (0.000) (0.038) (0.319)
Note: All regressions include controls for age, gender, education and Phnom Penh as in columns 1, 3, 5, 6, 7, 8, 9, 10, and 11 in Table 3. We assume that various birth cohorts born during the
KR period have unreliable memory of direct exposure to violence and recoded the value of their direct exposure measure as zero. P-values based on robust standard errors clustered at the district
level (68 districts) are reported in parentheses. *** significant at 1%, ** significant at 5%, * significant at 10%.
38
Table 6: Robustness to Post-Genocide Relocation
Money Burning Self-Reporting Dictator Trust Risk Index
Burn Dishonesty Take % Given % Sent % Returned % Invested Antisocial Prosocial
(1) (2) (3) (4) (5) (6) (7) (8) (9)
KR mortality rate (β1) -0.018 -0.688*** -0.249 -4.398 -10.130 -6.102 10.369 -8.461 -6.443
(0.162) (0.210) (0.247) (8.392) (11.538) (11.800) (8.338) (7.170) (9.792) Direct exposure (β2) -0.063 0.056 -0.126 -0.395 -5.086 -3.353 -4.950 -2.000 -3.119
(0.080) (0.059) (0.080) (4.969) (4.193) (3.117) (3.474) (2.425) (3.253)
KR mortality rate × Direct exp. (β3) 0.577*** 0.591** 0.584*** -20.389 0.657 4.751 24.028** 18.675** -4.242
(0.166) (0.259) (0.208) (15.694) (11.334) (13.999) (9.379) (7.069) (11.929)
Age 0.004 -0.003 -0.001 0.032 -0.005 -0.172 -0.186 -0.078 -0.049
(0.005) (0.003) (0.005) (0.303) (0.200) (0.163) (0.220) (0.090) (0.166) Education (years) 0.017*** -0.017*** 0.003 0.512 0.606* 0.300 0.929** 0.036 0.494*
(0.005) (0.006) (0.006) (0.387) (0.315) (0.220) (0.410) (0.137) (0.268)
Male -0.041 -0.091** -0.121*** 9.297*** 8.397*** 3.129** 8.682*** -2.149 6.858***
(0.051) (0.045) (0.039) (2.289) (2.373) (1.451) (2.566) (1.953) (1.418)
Phnom Penh -0.206** -0.172** -0.160 17.275*** 22.102*** 14.560*** 2.750 -7.178*** 18.215***
(0.093) (0.078) (0.103) (4.130) (3.900) (3.336) (4.679) (2.024) (3.070) Relocation 0.098 -0.029 -0.013 2.406 -0.488 -2.042 5.193* 0.467 0.218
(0.061) (0.047) (0.068) (2.930) (3.704) (2.455) (2.898) (2.177) (2.640)
R-squared 0.141 0.132 0.060 0.094 0.040 0.180 Observations 492 492 492 492 492 4920 492 492 492
KR mortality rate +
KR mortality rate × Direct exp. (β1+ β3) 0.559*** -0.097 0.335** -24.787* -9.473 -1.351 34.397*** 10.214** -10.685
(p-values) (0.000) (0.515) (0.013) (0.055) (0.435) (0.897) (0.000) (0.035) (0.285)
Note: All regressions include controls for age, gender, education and Phnom Penh as in columns 1, 3, 5, 6, 7, 8, 9, 10, and 11 in Table 3. Relocation is coded as one if the participant’s current
district of residence differs to the district of residence during the KR period or at birth (post-KR cohorts). Robust standard errors clustered at the district level (68 districts) are reported in
parentheses. *** significant at 1%, ** significant at 5%, * significant at 10%
.
39
Table 7: Robustness to the Exclusion of Education and Age Controls Money Burning Self-Reporting Dictator Trust Risk Index
Burn Dishonesty Take % Given % Sent % Returned % Invested Antisocial Prosocial
(1) (2) (3) (4) (5) (6) (7) (8) (9)
A. Excluding education control
KR mortality rate (β1) -0.007 -0.712*** -0.236 -3.905 -8.286 -4.507 10.948 -8.539 -5.174 (0.158) (0.212) (0.239) (7.570) (12.029) (11.920) (9.158) (6.837) (9.701)
Direct exposure (β2) -0.078 0.065 -0.124* -0.856 -5.130 -3.104 -5.911 -2.078 -3.247 (0.068) (0.061) (0.074) (4.708) (4.162) (3.244) (3.594) (2.288) (3.206)
KR mortality rate × Direct exp. (β3) 0.601*** 0.536** 0.592*** -19.177 2.339 5.720 26.163** 18.738*** -2.918
(0.158) (0.268) (0.211) (16.164) (11.721) (14.040) (9.955) (6.984) (12.292)
Age 0.002 0.001 -0.002 -0.061 -0.160 -0.273* -0.345** -0.081 -0.167
(0.004) (0.003) (0.004) (0.233) (0.201) (0.147) (0.162) (0.079) (0.136)
Male -0.015 -0.112** -0.118*** 10.052*** 9.110*** 3.385** 10.096*** -2.080 7.472***
(0.050) (0.045) (0.043) (2.346) (1.997) (1.458) (2.520) (1.967) (1.183) Phnom Penh -0.073 -0.242*** -0.156* 20.936*** 24.285*** 14.482*** 9.921** -6.728*** 20.389*** (0.071) (0.063) (0.093) (4.625) (4.502) (3.404) (4.204) (1.922) (3.704)
R-squared 0.133 0.125 0.057 0.070 0.039 0.171 Observations 492 492 492 492 492 4920 492 492 492
KR mortality rate +
KR mortality rate × Direct exp. (β1+ β3) 0.594*** -0.176 0.356*** -23.082* -5.947 1.213 37.111*** 10.199** -8.092
(0.000) (0.253) (0.004) (0.083) (0.641) (0.910) (0.000) (0.012) (0.431)
B. Excluding age and education controls
KR mortality rate (β1) -0.010 -0.713*** -0.233 -3.822 -8.070 -4.138 11.415 -8.430 -4.948
(0.156) (0.209) (0.241) (7.638) (12.075) (11.945) (9.154) (6.984) (9.752) Direct exposure (β2) -0.064 0.073 -0.141** -1.417 -6.595** -5.603** -9.069*** -2.816 -4.774*
(0.052) (0.055) (0.061) (4.045) (3.304) (2.640) (3.170) (2.153) (2.746)
KR mortality rate × Direct exp. (β3) 0.601*** 0.535** 0.593*** -19.142 2.431 5.877 26.362** 18.784** -2.822
(0.159) (0.267) (0.214) (16.044) (11.554) (13.672) (9.988) (7.127) (12.038)
Male -0.015 -0.112** -0.117*** 10.056*** 9.120*** 3.402** 10.118*** -2.075 7.482***
(0.050) (0.045) (0.043) (2.345) (1.976) (1.413) (2.444) (1.964) (1.162) Phnom Penh -0.076 -0.244*** -0.153 21.048*** 24.577*** 14.981*** 10.552** -6.581*** 20.694***
(0.069) (0.063) (0.095) (4.505) (4.560) (3.314) (4.243) (1.928) (3.616)
R-squared 0.133 0.124 0.053 0.065 0.038 0.169
Observations 492 492 492 492 492 4920 492 492 492
KR mortality rate +
KR mortality rate × Direct exp. (β1+ β3) 0.591*** -0.178 0.360*** -22.964* -5.639 1.739 37.777*** 10.354** -7.770
(p-values) (0.000) (0.245) (0.004) (0.082) (0.660) (0.871) (0.000) (0.010) (0.448)
Note: Robust standard errors clustered at the district level (68 districts) are reported in parentheses. *** significant at 1%, ** significant at 5%, * significant at 10%
40
Table 8: Robustness to the Inclusion of Risk Attitude Money Burning Self-Reporting Dictator Trust Risk Index
Burn Dishonesty Take % Given % Sent % Returned % Invested Antisocial Prosocial
(1) (2) (3) (4) (5) (6) (7) (8) (9)
KR mortality rate (β1) -0.055 -0.687*** -0.229 -6.602 -11.245 -6.801 8.264 -8.624 -7.909 (0.154) (0.221) (0.244) (8.327) (12.467) (12.525) (8.825) (7.100) (10.340)
Direct exposure (β2) -0.077 0.064 -0.135* 0.098 -4.105 -1.993 -5.721 -2.088 -2.197 (0.074) (0.057) (0.072) (4.697) (3.983) (3.029) (3.448) (2.207) (3.048)
KR mortality rate × Direct exp. (β3) 0.565*** 0.586** 0.629*** -24.128 -3.025 0.616 23.501** 18.702*** -8.182
(0.164) (0.261) (0.214) (16.022) (11.144) (13.660) (9.256) (6.758) (11.885)
Age 0.006 -0.003 -0.001 0.086 0.004 -0.185 -0.104 -0.071 -0.028
(0.004) (0.003) (0.004) (0.294) (0.192) (0.186) (0.212) (0.080) (0.172)
Education (years) 0.018*** -0.018*** 0.005 0.391 0.445 0.099 0.980** 0.044 0.332
(0.005) (0.006) (0.005) (0.404) (0.339) (0.267) (0.399) (0.124) (0.298) Male -0.036 -0.099** -0.106*** 8.074*** 6.963*** 1.398 8.900*** -2.101 5.383***
(0.050) (0.046) (0.038) (2.420) (2.517) (1.471) (2.451) (1.817) (1.482)
Phnom Penh -0.148* -0.196*** -0.161 17.852*** 20.873*** 12.241*** 5.883 -6.877*** 17.366*** (0.086) (0.073) (0.100) (4.256) (4.000) (3.158) (4.371) (2.210) (3.213)
Risk - % Invested -0.000 0.001 -0.002*** 0.149*** 0.159*** 0.185*** -0.003 0.167***
(0.001) (0.000) (0.001) (0.042) (0.030) (0.032) (0.021) (0.026) R-squared 0.164 0.155 0.098 0.088 0.039 0.228
Observations 492 492 492 492 492 4920 492 492 492
KR mortality rate +
KR mortality rate × Direct exp. (β1+ β3) 0.510*** -0.101 0.400*** -30.730** -14.270 -6.185 31.765*** 10.078** -16.091
(p-values) (0.001) (0.494) (0.000) (0.021) (0.277) (0.585) (0.000) (0.024) (0.132)
Note: All regressions include controls for age, gender, education and Phnom Penh as in columns 1, 3, 5, 6, 7, 8, 9, 10, and 11 in Table 3. Robust standard errors clustered at the district level (68
districts) are reported in parentheses. *** significant at 1%, ** significant at 5%, * significant at 10%
41
NOT FOR PUBLICATION
Appendix Table
Table A1: Estimates of effects of exposure to genocide on prosocial and antisocial behaviour and risk
Money Burning Self-reporting Dictator Trust Risk Index
Dependent variable: Burn Burn Dishonest Dishonest Take % Given % Sent % Returned % Invested Antisocial
Preferences
Prosocial
Preferences
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
KR mortality rate (β1) -0.306 -0.317* -0.933*** -0.960*** -0.487** -12.369 -5.516 -8.608 23.033** -13.163 -8.113 (0.194) (0.180) (0.240) (0.246) (0.221) (10.916) (14.458) (16.740) (8.376) (8.327) (13.502)
Direct exposure (β2) -0.098 -0.112 0.048 0.032 -0.135 -1.507 -3.554 -5.751 -7.037 -1.618 -3.865 (0.135) (0.140) (0.070) (0.071) (0.095) (6.288) (3.396) (4.574) (3.893) (3.038) (3.228)
KR mortality rate × Direct exp. (β3) 0.923*** 0.957*** 0.809*** 0.862*** 0.711*** -8.591 -11.727** 5.093 15.316* 21.241** -4.117
(0.243) (0.232) (0.288) (0.286) (0.159) (13.344) (4.041) (15.162) (8.351) (6.664) (9.202)
Age -0.001 -0.001 -0.003 -0.003 -0.004 -0.336 0.115 -0.045 -0.074 -0.130 -0.087
(0.007) (0.007) (0.005) (0.005) (0.008) (0.349) (0.249) (0.168) (0.329) (0.108) (0.185)
Education (years) 0.014* 0.014* -0.022*** -0.025*** 0.001 0.204 0.763* 0.596 0.885 -0.088 0.541
(0.007) (0.007) (0.009) (0.009) (0.009) (0.442) (0.415) (0.347) (0.639) (0.161) (0.367)
Male 0.018 0.018 -0.117* -0.131** -0.108** 9.494*** 8.458** 3.756** 8.282** -1.641 7.203***
(0.074) (0.074) (0.063) (0.055) (0.055) (2.112) (2.995) (1.399) (2.658) (2.525) (1.092)
Phnom Penh -0.268 -0.273 -0.092 -0.155 0.000 16.533** 15.397*** 13.121 16.330 -12.523*** 14.978*
(0.337) (0.344) (0.205) (0.205) (.) (7.003) (3.425) (16.813) (10.423) (2.750) (7.728)
Advantaged player -0.037
(0.052)
Maximum number of correct answers -0.177**
(0.074)
R-squared 0.088 0.055 0.032 0.086 0.033 0.093
Observations 316 316 316 316 312 316 316 3160 316 316 316
KR mortality rate +
KR mortality rate × Direct exp. (β1+
β3) 0.617** 0.64** -0.124 -0.098 0.224** -20.96** -17.243 -3.515 38.349*** 8.078** -12.23
(p-values) (0.004) (0.004) (0.266) (0.383) (0.046) (0.011) (0.257) (0.801) (0.000) (0.013) (0.252)
Note: This Table reports results for the same set of variables as in Table 3. However, the sample includes individuals who have been in the same district since the beginning of
the KR regime. The estimation method, control variables are all the same as in Table 3.
42
NOT FOR PUBLICATION
Appendix A: Experimental Game Design and Procedure
The following section describes the experimental games with the endowment and participation
fees used in the rural areas. One of the first four tasks was randomly chosen for payment
purposes, in addition to the earnings in the self-reporting task.
Game 1: Trust
We use the standard trust game protocol to measure trust and trustworthiness. Each participant
plays both as player 1 (sender) and player 2 (receiver). The sender receives an endowment of
10,000 riel22, while the receiver is endowed with 0 riel. In the first stage, all participants are
senders and can send any positive amount x {0, 1,000, 2,000, …, 9,000, 10,000} to the
anonymous receiver, knowing that the experimenter triples the amount sent and that the receiver
receives an amount of 3x. In the second stage, all participants play as receivers. We aimed to
minimize the logistical issues in the field, so the receiver was not informed of the amount sent
by the sender. Thus, the receiver decides on an amount y {3x} to return to the sender for all
the corresponding amounts the receiver might receive. The sender is also not informed of the
amount sent back by the receiver unless the game is selected for the final payment. If the trust
game and the sender’s role are selected for the final payment, all participants receive (10,000 –
x + y). Otherwise, all participants receive (3x – y) if the receiver’s role is chosen.
Game 2: Dictator
In the dictator game, there were two stages: 1) giving; and 2) giving or taking. Each participant
plays as both player 1 (dictator) and player 2 (recipient). All participants receive an endowment
of 10,000 riel. The dictator receives an additional endowment of 10,000 riel, while the recipient
does not receive the additional endowment.
1) In dictator game giving, the dictator is asked to decide how much of the additional
endowment to give to the recipient. The dictator can transfer any positive amount x {0, 1,000,
2,000, …, 9,000, 10,000} to the anonymous recipient. The recipient must simply accept it and
is only informed of how much the dictator sends if the game is selected for the final payment.
All participants receive (10,000 + 10,000 – x) if the dictator game, part 1, and the dictator’s role
are selected for the final payment. However, if the dictator game, part 1, and the recipient’s role
are chosen, the payoffs for all participants are (10,000 + x).
2) In dictator game giving or taking, the dictator can send the additional endowment to
other players or take other player’s initial endowment. This means that the dictator can send a
negative or positive amount x {–1,000, –2,000, …, –10,000, 0, 1,000, 2,000, …, 10,000}. As
in part 1, the recipient is only told the amount the dictator sends or takes if the game is selected
for the final payment. The payoffs of all participants are (10,000 + 10,000 –/+ x) depending on
the dictator’s decision to take or give if the dictator game, part 2, and the dictator’s role are
selected and (10,000 –/+ x) if the dictator game, part 2, and the recipient’s role are selected.
22 The exchange rate was 1USD = 4037.5Riel (2014).
43
Game 3: Risk
We use a simple risk game which involves a 50% chance of winning or losing. Each participant
receives 10,000 riel and can invest any positive amount x {0, 1,000, 2,000, …, 9,000, 10,000}
in a risky business. The investment yields triple the amount invested with 50% probability and
0 with 50% probability. The outcome is decided by tossing a coin. If the coin shows heads, the
investment is successful, and all participants gain (10,000 – x + 3x). If the coin shows tails, the
payoff for all players is (10,000 – x + 0).
Game 4: Money Burning
In the money burning game introduced by Zizzo and Oswald (2001), each player is given an
opportunity to pay a fee to reduce the income of the other player. We use a simpler two player
version of this game.
All participants receive 20,000 riel. Half of them receive an additional amount called a
gift. Those who have odd identification (ID) numbers receive a gift of 5,000 riel, and those with
even ID numbers do not receive any gift. The gift is known to all participants. Participants
simultaneously decide how much of the other player’s total endowment to eliminate or not to
eliminate any. Participants have to pay from their own endowment to eliminate the other
player’s endowment. The fee incurred for eliminating other’s endowment is charged at three
levels: 5%, 10%, and 20% of the amount a player wants to eliminate of the other player’s
endowment. We study different costs of eliminating to test whether the cost has any influence
on an individual’s behaviour. In the payment stage, if this game is chosen, the odd-numbered
participants chose an even-numbered partner by randomly selecting an even ID number, and
vice versa.
Game 5: Self-reporting
In the self-reporting game, we aim to measure dishonesty using an individual-level decision-
making environment. We design a simple self-reporting task with pictures instead of games
with numerical or word tasks to accommodate the low literacy level in Cambodia. The game
involves finding the picture of a star from a sheet of 10 tables which each has 9 images (see
Appendix B). Each participant is given an envelope containing a sheet of 10 tables and is
instructed to find the stars within 1 minute.
To ensure that considerable and different opportunities for cheating, not all of the 10
tables have a star. We design 2 different sheets: a sheet with 7 stars in the 10 tables and a sheet
with only 4 stars in the 10 tables. These maximum numbers are not known to the participants.
The maximum number of 4 or 7 stars for each sheet allows considerable scope for cheating,
even for top performers. In rural areas, participant can earn 1,000 riel23 for each star found.
Participants record the total number of stars they find at the end of the sheet, place the sheet
back into the envelope, and pay themselves for this task from a small envelope containing ten
1,000 riel notes given to them at the beginning of the task. Participants place any remaining
money in the small envelope, seal it, and leave it on their desk for the experimenters to collect.
The envelopes are not opened until the experimental session is completed. To reduce scrutiny
bias, the experimenters leave the room while participants perform this task.
23 It was 2,000 riel in Phnom Penh.
44
Experimental Procedure
All the tasks were conducted with paper and pen. Clear instructions with tables and diagrams
in Khmer were provided to all participants. Before starting the experiment, participants were
randomly assigned to 1 of 3 separate rooms in a local school.24 On average, there were 24
participants per session. One session was smaller (14 participants) because of the small size of
the room available and 3 sessions were larger (30–32 participants). We ran 3 sessions (rooms)
simultaneously in Phnom Penh and in each district in Kampong Cham to reduce the spillover
effects between sessions.
Participants played with other participants in the same session. They were informed that
their partner was another participant in the session and selected their partner during the payment
stage by choosing the ID number of another participant in the same session. Participants were
also informed that they would be paid for 1 of the first 4 tasks picked at random, plus their
earnings in task 5 and participation fees of 20,000 riel in Phnom Penh and 10,000 riel in the
rural areas. At the end of the entire experiment, the experimenter rolled a dice in front of the
participants to determine for which game participants were paid. Participants did not receive
any feedback between the tasks or on the tasks not chosen for the final payment.
The games were conducted in two different orders. Game order 1 followed the sequence
of trust, dictator, risk, money burning, and self-reporting games. This order was followed in
odd-numbered districts (4 of 7 districts). In game order 2, we used the sequence of risk, money
burning, trust, dictator, and self-reporting games in even-numbered district (3 districts). We
altered the order of the games mainly to test whether participating in the antisocial games first
might influence participants’ behaviour differently.25 Participants had to make decisions in a
booth during each game, except for the self-reporting game. An experimenter was in the booth
to assist participants if they could not read or write.
24 In Phnom Penh, we conducted the experiment at a research institute: the Cambodia Development Resource
Institute. 25
Our results are robust when we control for game order instead of experimental district fixed effects. The self-
reporting task was always conducted last as participants paid themselves in this task and the amount they earned
in this task could potentially influence their decisions in other tasks if this task were conducted before the others.
45
Appendix B: Instructions and Decision Sheets
Instructions for Participants
Dear Participant,
Thank you for participating in this experiment.
There will be two main parts of today’s experiment: performing some tasks and answering the
questionnaires. There are five tasks; and we are going to ask you to make some decisions in
each task. You will have an opportunity to earn a considerable amount of money depending on
the decisions you and everyone else make during each task. You will be paid for ONE of the
first four tasks plus your earnings in the Task 5. We will invite one of the participants in the
room to draw 1 ball out of the box of 4 balls (representing the four tasks). The number of the
ball defines which game will be paid to EVERYONE in the room at the end of the entire
experiment. In addition, you will also be given an additional 10,000 riels for your participation
in this research.
Before we start, we will read the consent form. If you agree to participate in this experiment,
you are required to sign the form. Then we will collect the signed form from you. If you do not
agree, you can leave and will not receive the 10,000 riels participation fee.
The experiment will last approximately 2-2.5 hours depending on your cooperation. You can
decide not to participate at any time if you think that you are not happy with the experiment.
All decisions that you make in the tasks and responses that you provide during the interview
will be completely anonymous and will be recorded only by anonymous ID number.
You will be invited to randomly select a small envelope before we start. In the envelope, there
is a paper with your ID number. Please do not remove the ID number from the envelope and
keep this envelope (ID number) with you all the times and do not show this ID number to
anyone during and after this experiment. You will need to present your ID number to the
research assistant who will collect the decision sheet and will interview you after you complete
all the five tasks and to the cashier to receive your payment at the end of the entire experiment.
We are about to begin the first task. We will explain the task and give you some examples.
When we finish reading the instruction, you will have a chance to ask questions. During
performing each task, you also can raise your hand if you are unclear and one of our team will
come by to help you. It is very important that you understand how to do in each task. If you do
not understand, you will not be able to participate effectively. While you are waiting for other
participant to make his/her decision during this experiment, please wait at your seat and do not
communicate with other participants. Also, please do not share your decision with other
participants during this experiment to ensure the confidentiality.
If you are ready, then we will proceed. Please follow the experimenter.
46
Instructions for Task 1 (Trust Game)
In Task 1, there will be two stages as shown in this diagram. You will be asked to make a
decision involving money. This task will be played by pairs of individuals. Each pair is made
up of a “Sender” and a “Receiver”. Each of you will be paired randomly with another participant
in this room. You will not be told who you are matched with during or after the experiment.
You will play as both Sender and Receiver in the Task 1. In Stage 1, you will play as “Sender”
and will be paired with one of the participants in this room. In Stage 2, you will play as
“Receiver” and will be paired with different partner in Stage 1.
If this task is chosen for payment at the end of the experiment, we will decide which role
(Sender or Receiver) for the payment for EVERYONE in the room by tossing a coin. We will
invite one person from your group to come out here and toss the coin. If the head shows up,
the role as Sender will be chosen for the payment. If the tail shows up, the role as Receiver
will be chosen for the payment.
Stage 1: Sender
In the first stage, each of you plays as a “Sender” and has been allocated 10,000 riels. No money
will be given at this point. All actual payments will be made at the end of the experiment if this
task is chosen for the final payment when the experiment finishes.
Your decision: Decide how much of your endowment (10,000 riels) you would allocate
to your partner (Receiver) who is also the participant in this session. You can decide to
keep all money for yourself or allocate some or all of it to your partner. The experiment
will multiply what you gave by three and pass the money to your partner. Then, in the
second stage, your partner will have an opportunity to keep all of the money you sent
to him/her or to return some or all of it to you.
As shown in the decision sheet #1 that we just distribute to you, you can choose ONLY
one option.
For example, if you choose to allocate 5000 riels to your partner, your partner will get
15,000 riels (= 5000 riels x 3 times). Your partner will make decision to send the money
back to you in the second stage. If he/she sends back 4000 riels. Then you will have
9000 riels (=10,000 riels – 5000 riels + 4000 riels) to bring home if this game is chosen
for the final payment. Your partner will bring home 11,000 riels (=(5000 riels x 3 times)
– 4000 riels).
If you decide to give nothing (zero) to your partner, the first stage for you ends. Your
pay-off from this stage is 10,000 riels and your partner’s pay-off is 0 riel.
Please note that these are only examples. The actual decisions are up to you.
47
Procedure:
You will be asked to go to the private booth at the corner and bring with you the “ID
number” and the “decision sheet”. In the booth, you decide how much money you want
to allocate to your partner by choosing one of the options in the decision sheet.
Then, the research assistant will ask you the amount of money you want to allocate to
your partner to confirm your decision, write down your ID number on the decision sheet
and put the decision sheet inside a box. Then you can go back to your seat. Please keep
in mind that you are not allowed to discuss or let other participants know about your
decision.
Do you have any question? If you are ready, we will proceed. Please remain seated until we
invite you to the booth.
Stage 2: Receiver
In the second stage, each of you has been assigned as “Receiver”. Each of you is now paired
randomly with other participant in this room. Remember your partner in this stage is different
from the first stage. You will not be told who you are matched with during or after this task.
Your partner in the first stage has made a decision of how much of his/her endowment
(10,000 riels) he/she is willing to allocate to you. He/she was told that the amount he
would give will be tripled and given to you. So you would have an opportunity to return
money that you receive.
Your decision: Indicate how much of the money you are willing to return conditional
on how much you received. To be specific, how much you are going to return to your
partner, if you received 3000 riels, 6000 riels, 9000 riels, 12,000 riels, 15,000 riels,…,
30,000 riels. This means you have to complete all these blanks as shown in this decision
sheet #2. You can keep all the money, return all or some of it.
For example:
o If your partner in Stage 1 sent you 3000 riels (already triple), you decide to
return to your partner 0 riel.
o If your partner in Stage 1 sent you 12,000 riels (already triple), you decide to
send back 8000 riels.
o If your partner in Stage 1 sent you 30,000 riels (already triple), you decide to
return to your partner 10,000 riels.
Please note that these are only examples. The actual decisions are up to you.
48
Procedure:
You will be asked to go to the private booth at the corner and bring with you the “ID
number” and the “decision sheet”. In the booth, you indicate how much you will return
to your partner (please fill in all blanks).
The research assistant will ask you the amount you want to return in each blank to
confirm your decision and write down your ID number in the decision sheet, and put
the decision sheet inside a box. The research assistant will assist you to write the amount
based on your decision in each blank if needed. Then you can go back to your seat.
Please keep in mind that you are not allowed to discuss or let other participants know
about your decision.
Do you have any question? If you are ready, we will proceed. Please remain seated until we
invite you to the booth.
49
Decision Sheet of Trust Game: Sender’s Role
DECISION SHEET #1
PARTICIPANT ID: _______________
TASK 1
STAGE 1
Please decide: “how much would you like to allocate to your partner?”
Please select only one option from the following:
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
8,000
9,000
10,000
50
Decision Sheet of Trust Game: Recipient’s Role
DECISION SHEET #2
PARTICIPANT ID: _______________
TASK 1
STAGE 2
The amount in column A is the amount you received from your partner. The amount is already tripled.
Please decide: “how much would you like to return to your partner?” Please write down your decision in
each row in column B according to the amount you received from your partner in column A.
A B
Amount received from your partner (already tripled) Amount you wish to send
back to your partner
3,000
__________________Riel
6,000
__________________Riel
9,000
__________________Riel
12,000
__________________Riel
15,000
__________________Riel
18,000
__________________Riel
21,000
__________________Riel
24,000
__________________Riel
27,000
__________________Riel
30,000
__________________Riel
51
Instructions for Task 2 (Dictator Game)
There will be two parts in Task 2. You will be asked to make a decision involving money. This
task will also be played by pairs of individuals. Each pair is made up of a “Sender” and a
“Receiver”.
If Task 2 is chosen for payment at the end of the experiment, we will decide which part (PART
I or PART II) for the payment for EVERYONE in the room by tossing a coin. We will invite
one person from your group to come out here and toss the coin. If the head shows up, PART I
will be chosen for the payment. If the tail shows up, PART II will be chosen for the payment.
Then, we will decide which role (Sender or Receiver) in the chosen PART above for the
payment for EVERYONE in the room by tossing the coin again. We will invite another person
from your group to come out here and toss the coin. If the head shows up, the role as Sender
will be chosen for the payment. If the tail shows up, the role as Receiver will be chosen for the
payment.
Part I: Giving
Each of you is assigned the role as a “Sender”. You are now paired randomly to anyone in this
room. You will not be told who you are matched with during or after the experiment.
Both you and your partner have been allocated 10,000 riels in this part of the
experiment. In addition, only you as “Sender” have been allocated an additional 10,000
riels. No money will be given at this point. All actual payments will made at the end of
the experiment if this task is chosen for the final payment when the experiment finishes.
Your decision: Decide how much of your additional endowment (10,000 riels) to
allocate to your partner. You can choose any amount from as shown in the decision
sheet #3.
Your partner will not make any decision. He/she will take home whatever amount you
allocate to him/her plus his/her initial 10,000 riels allocation, and you are as “Sender”
will take home whatever amount you keep for yourself plus your initial 10,000 riels
allocation.
For example, you choose to allocate 4000 riels to your partner. Then your partner will
go home with 14,000 riels (=10,000 riels + 4000 riels) and you will go home with 16,000
riels (10,000 riels +10,000 riels – 4000 riels).
Please note that this is an example only. The actual decision is up to you.
Procedure:
You will be asked to go to the private booth at the corner and bring with you the “ID
number” and the “decision sheet”. In the booth, you decide how much money you want
to allocate to your partner by choosing ONLY one option in the decision sheet #3.
Then, the research assistant will ask you the amount of money you want to allocate to
your partner to confirm your decision, write down your ID number on the decision sheet
and put the decision sheet inside a box. Then you can go back to your seat. Please keep
52
in mind that you are not allowed to discuss or let other participants know about your
decision
Do you have any question? If you are ready, we will proceed. Please remain seated until we
invite you to the booth.
Part II: Giving or Taking
In the second part, there is the same procedure as in Part I. You are assigned the role as a
“Sender”. You are now paired randomly to a new partner from anyone in this room. You will
not be told who you are matched with during or after the experiment. Please keep in mind that
your partner in this part is different from your partner in Part I.
You and your new partner have been allocated 10,000 riels. In addition, only you as
“Sender” have been allocated an additional 10,000 riels. No money will be given at this
point. All actual payments will be made at the end of the experiment if this task is chosen
for the final payment.
Your decision: Decide how much of your additional endowment (10,000 riels) to
transfer to your partner. Moreover, you can also transfer a negative amount. This means
that you can take up to 10,000 riels from your partner. You can choose any amount from
the options in the decision sheet #4.
The minus amount means you take your partner’s endowment instead of dividing your
endowment to your partner.
For example, if you decide to choose – 6000 riels, it means you take 6000 riels from
your partner’s endowment. So your partner will take home 4000 riels (=10,000 riels –
6000 riels) and you will have 26,000 riels (=10,000 riels + 10,000 riels + 6000 riels).
If you decide to choose 5000 riels to transfer to your partner, he/she will take home
15,000 riels (=10,000 riels + 5000 riels) and you will take home 15,000 riels (=10,000
riels + 10,000 riels – 5000 riels)
Please note that these are only examples. The actual decisions are up to you.
Procedure:
You will be asked to go to the private booth at the corner and bring with you the “ID
number” and the “decision sheet”. In the booth, you decide how much money you want
to allocate to OR take from your partner by choosing one of options in the decision sheet
#4.
Then, the research assistant will ask you the amount of money you want to allocate to
OR take from your partner to confirm your decision, write down your ID number on the
decision sheet and put the decision sheet inside a box. Then you can go back to your
seat. Please keep in mind that you are not allowed to discuss or let other participants
know about your decision.
Do you have any question? If you are ready, we will proceed. Please remain seated until we
invite you to the booth.
53
Decision Sheet of Dictator Game: Giving
DECISION SHEET #3
PARTICIPANT ID: _______________
TASK 2
PART 1
Please decide: “how much would you like to allocate to your partner (using additional money
10,000 Riel)?” Please select only one option from the following:
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
8,000
9,000
10,000
54
Decision Sheet of Dictator Game: Giving or Taking DECISION SHEET #4
PARTICIPANT ID: _______________
TASK 2
PART II
Please decide: “how much would you like to transfer to your partner or how much would you like to take from your partner?”
Please select only one option from the following:
- 10,000 minus
- 9,000 minus
- 8,000 minus
- 7,000 minus
- 6,000 minus
- 5,000 minus
- 4,000 minus
- 3,000 minus
- 2,000 minus
- 1,000 minus
0
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
8,000
9,000
10,000
55
Instructions for Task 3 (Risk Game)
In Task 3, once again, you will be asked to make a decision involving money in this Task.
You have been allocated 10,000 riels. No money will be given at this point. All actual
payments will be made at the end of the experiment if this task is chosen for the final
payment..
You have an opportunity to invest all, some or none of these 10,000 riels.
There is 50% chance of successful investment and 50% chance of unsuccessful
investment. If the investment succeeds, you will receive 3 times the amount of money
you invested. If the investment fails, you will lose the amount you invested.
We will decide the outcome of the investment by tossing a coin at the end of the entire
session if this task is chosen for payment. We will invite one person from your group to
come out here and toss the coin in front of all of you. If the coin shows head,
EVERYONE in this room get 3 times of the amount that each participant invested. But
if the coin shows tail, EVERYONE in this room will get nothing.
For examples:
- If you choose to invest nothing, you will get 10,000 riels if this task is chosen for
payment at the end of the experiment.
- If you choose to invest all of the 10,000 riels, you get 30,000 riels (=10,000 riels x 3
times) if head shows up or nothing (zero riels) if it’s tail.
- If you choose to invest 4000 riels, you receive 18,000 riels (= [4000 riels x 3 times] +
6000 riels) if the head shows up and 6,000 riels (=10,000 riels – 4000 riels) if the tail
shows up.
Please note that these are only examples. The actual decisions are up to you.
Procedure:
You will be asked to go to the private booth at the corner and bring with you the “ID
number” and the “decision sheet”. In the booth, you choose how much you want to
invest from the decision sheet #5.
Then, the research assistant will ask you the amount of money you want to invest to
confirm your decision, write down your ID number on the decision sheet and put the
decision sheet inside a box. Then you can go back to your seat. Please keep in mind that
you are not allowed to discuss or let other participants know about your decision
Do you have any question? If you are ready, we will proceed. Please remain seated until we
invite you to the booth.
56
Decision Sheet of Risk Game
DECISION SHEET #5
PARTICIPANT ID: _______________
TASK 3
Please decide: “how much would you like to invest in a risky business?” Please select only one option
from the following:
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
8,000
9,000
10,000
57
Instructions for Task 4 (Money Burning Game)
In Task 4, you are now paired randomly to a new partner from anyone in this room. You will
not be told who you are matched with during or after the experiment.
You and your partner have been allocated 20,000 riels.
We divide you into two groups: Group 1 has Odd ID number and Group 2 has Even ID
number.
The person who has Odd ID number receives a gift of 5000 riels.
We will pair the person who has Odd ID number with the person who has Even ID
number. Please remember that you will not be told who you are matched with during or
after the experiment.
Your and your partner’s incomes are public knowledge, meaning that you can know
how much your partner’s income is, and vice versa your partner will learn about your
income too.
Your decision: Decide how much you would like to reduce your partner’s total income.
But you have to pay from your income for reducing other person’s income. There are
three prices of reducing other person’s income: 5%, 10% and 20% of the amount you
want to reduce.
The gray box in the “decision sheet #6/7” is your and your partner’s total income. The
price of reducing is at 5% in Case 1, 10% in Case 2, and 20% in Case 3. You have to
choose ONLY ONE option in Case 1, Case 2 and Case 3.
For example: you have Odd ID number, you receive gift (5000 riels) and your partner
does not. So your total income is 25,000 riels and your partner’s total income is 20,000
riels.
In Case 1 (5% price of reducing), you decide to reduce 10,000 riels of your partner’s
income. Thus, you will have to pay from your income by 500 riels (= 5% * 10,000 riels).
Your partner decides to reduce 20,000 riels of your income; hence, he/she will have to
pay from his/her income by 1000 riels (= 5% * 20,000 riels).
Your income will remain 4500 riels (= 25,000 riels – 500 riels – 20,000 riels) and your
partner’s income will be 9000 riels (= 20,000 riels – 1000 riels – 10,000 riels).
Please note that these are only examples. The actual decisions are up to you.
Other participant has been told to make the same decision. This means that your partner
can do the same thing as you are being told.
Both you and your partner make decision simultaneously, but only the choice of one
player (your partner or your choice) will be randomly selected for the payment if the
task is chosen for the final payment. Then, we will also randomly select which
case/price of reducing (5%, 10% or 20%) for the final payment.
58
Procedure:
You will be asked to go into the private booth at the corner and bring with you the “ID
number” and the “decision sheet”. In the booth, you decide how much money you want
to reduce other person’s income in Case 1, Case 2 and Case 3. Then, you estimate how
much do you think your partner decides to reduce your income.
Next, the research assistant will ask you the amount of money you want to reduce other
person’s income in each case to confirm your decision, write down your ID number on
the decision sheet and put the decision sheet inside a box. Then you can go back to your
seat. Please keep in mind that you are not allowed to discuss or let other participants
know about your decision.
Do you have any question? If you are ready, we will proceed. Please remain seated until
we invite you to the booth.
59
Decision Sheet of Money Burning Game
DECISION SHEET #6
(For ODD ID number)
PARTICIPANT ID: _______________
TASK 4
YOU Your Partner
Initial endowment: 20,000 Riel 20,000 Riel
Gift: 5,000 Riel 0 Riel
Total income:
25,000 Riel
20,000 Riel
Case 1: At 5% price of reducing other person’s income
How much do you want to reduce other
person’s income?
Please select only one option from the
following:
if the price for eliminating is:
0 0
5,000 250
10,000 500
15,000 750
20,000 1,000
60
Case 2: At 10% price of reducing other person’s income
How much do you want to reduce other
person’s income?
Please select only one option from the
following:
if the price for eliminating is:
0 0
5,000 500
10,000 1,000
15,000 1,500
20,000 2,000
Case 3: At 20% price of reducing other person’s income
How much do you want to reduce other
person’s income?
Please select only one option from the
following:
if the price for eliminating is:
0 0
5,000 1,000
10,000 2,000
15,000 3,000
20,000 4,000
61
Decision Sheet of Money Burning Game
DECISION SHEET #7
(For EVEN ID number)
PARTICIPANT ID: _______________
TASK 4
YOU Your Partner
Initial endowment: 20,000 Riel 20,000 Riel
Gift: 0 Riel 5,000 Riel
Total income:
20,000 Riel
25,000 Riel
Case 1: At 5% price of reducing other person’s income
How much do you want to reduce other
person’s income?
Please select only one option from the
following:
if the price for eliminating is:
0 0
5,000 250
10,000 500
15,000 750
20,000 1,000
25,000 1,250
62
Case 2: At 10% price of reducing other person’s income
How much do you want to reduce other
person’s income?
Please select only one option from the
following:
if the price for eliminating is:
0 0
5,000 500
10,000 1,000
15,000 1,500
20,000 2,000
25,000 2,500
Case 3: At 20% price of reducing other person’s income
How much do you want to reduce other
person’s income?
Please select only one option from the
following:
if the price for eliminating is:
0 0
5,000 1,000
10,000 2,000
15,000 3,000
20,000 4,000
25,000 5,000
63
Instructions for Task 5 (Self-reporting Game)
In Task 5, each participant will be given a large envelope and you will find a sheet of 10
matrices like the one below:
You will get different sheet of matrices.
Your role is to tick in the box if you see a “star” in the matrix.
There are 10 matrices and you have 1 minute for this Task. You will get 1000 riels
(USD0.25) for each “star” you found.
However, some of the matrices have no “star”.
When you find the matrix that does not have “star”, please mark in the box.
You will not receive the money for the matrix that does not have “star”. But you will
receive money based on number of star you found.
Please write down the total number of star you found in this box. For example, if you
find six stars, please write 6.
64
Total : ………… 6 ………
On your table, there is a small envelope containing ten 1000 riel notes.
Please pay yourself for this task based on the total number of stars you found.
When you finish, please put the matrix sheet and any remaining money in the small
envelope in this box.
Our team will leave the room for 1 minute.
Please note that we will not open the matrix sheet and count the remaining money until
you leave the room (the session finishes).
Do you have any question? If you are ready, we will proceed.
65
Decision Sheet of Self-reporting Game
TASK 5
- Please tick in the box if you find a star in the matrix.
- Please cross in the box if you do not find star in the matrix.
Example:
Total number of stars found:
…………………………
Appendix C: Construction of the District Level Morality Rates
We use geographic variation in KR mortality rates across districts to examine the impact of the
genocide on children’s outcomes. We construct KR mortality rates using information sourced
from the CGD and Census 1998. The CGD includes a district identifier of each KR mass
gravesite and the estimated number of bodies at each mass grave.1 Some graves have minimum
and maximum estimates of bodies and, in such situations, we use the average of the two
estimates to construct district-level KR-related deaths. To obtain our preferred measure of KR
mortality rates, we divide the estimated deaths under the KR in a district (based on information
in the CGD) by the sum of the estimated deaths under the KR and the number of individuals
who were in the district at the start of the KR regime (in 1975) and were still alive in 1998
(based on Census 1998 data).2 As we do not have information on the number of individuals
who survived the KR regime, but died between 1980 and 1998 at the district level, the estimated
KR mortality rates are noisy. Districts in five provinces (Kaoh Kong, Preah Vihear, Otdar Mean
Chey, Krong Kaeb, and Krong Pailin) are excluded from the analysis because no information
on the estimated deaths under the KR regime is available in the CGD.
1 The CGD was initially developed by Yale University and has been updated by the Documentation Center of
Cambodia (DC-Cam). We use data from both sources.
See http://www.yale.edu/cgp/ and http://www.d.dccam.org/Database/Index1.htm for details on the Cambodian
Genocide Program and DC-Cam database. 2 Living persons who were born between 1975 and 1979 are allocated to their districts of birth and included in the
denominator.
1
Appendix D: Directly Exposed and Indirectly Exposed Groups
Panel B of Table 1 shows that, in the trust game, individuals directly exposed to genocide send
on average less of their endowment to the other players (27.3% versus 33.4%; p = 0.018) and
return less (28.1% versus 32.8%; p = 0.000) than those indirectly exposed. The directly exposed
group also shares less of the endowment in the giving part of the dictator game, compared to
the indirectly exposed group (21.1% versus 26.4%; p = 0.023). In the dictator game giving or
taking, each individual has 3 options: (1) take some or all of the other player’s endowment; (2)
do not give to or take from the other player; and (3) give some or all of one’s endowment to the
other player. Overall, 33.7% of the directly exposed group (66 out of 196) chose option 1,
compared to 35.8% of individuals (106 out of 296) in the indirectly-exposed group. The figures
are similar for options 2: 31.6% for the directly exposed group and 27% for the indirectly
exposed group. However, 34.7% of the directly exposed group chose option 3, compared to
37.2% of the indirectly exposed group. We define a simple binary variable as equal to 1 if the
individual takes some or all of the other player’s endowment (option 1); otherwise, it is equal
to 0. Panel B of Table 1 and Figure A1 show that directly exposed individuals do not differ
significantly from indirectly exposed individuals in the dictator game taking.
[Appendix Figure A1]
Regarding risk-taking behavior, individuals directly exposed to genocide invest less than
indirectly exposed individuals (45.2% versus 47.3%) but the difference is not significant
statistically (p = 0.402). Figure A1 also shows the differences in the means in the anti-social
games. In the money burning game, we measure the burning decision by the choice to reduce
(burn) the other player’s money for at least 1 of the 3 prices (costs) of burning (5%, 10%, and
20%). More participants in the directly exposed group (51%) than the indirectly exposed group
(43.9%) choose to eliminate the other player’s money for at least 1 of the 3 prices. In the self-
reporting game, 41% of the directly exposed group takes extra money, compared to 24% of the
indirectly exposed group (panel B, Table 1). This difference is strongly statistically significant
(p = 0.000), suggesting that exposure to genocide can lead to dishonest behavior.
2
Figure A1: Percentage points of mean differences in experimental outcomes between
directly exposed and indirectly exposed groups
Note: The directly exposed group and the indirectly exposed group are defined in section 4.3 of the paper. The
burn variable equals 1 if the player decides to reduce (burn) the other player’s money for at least 1 of the 3 prices
and 0 if player decides not to burn any money. The dishonest variable equals 1 if the player takes extra money to
which he or she is not entitled and 0 otherwise. The giving or taking variable equals 1 if the player decides to take
some or all of the other player’s endowment and 0 otherwise.
p=0.123
p=0.000
p=0.627
p=0.023 p=0.018p=0.000
p=0.402
18
16
14
12
10
8
6
4
2
0
-2
-4
-6
-8
Burn
DishonestTake
% Given
% Sent
% Returned
% Invested