University of Southern California Dissertations and Theses

Direct and indirect predictors of traumatic stress and distress in orphaned survivors of the 1994 Rwandan Tutsi genocide

(USC Thesis Other)

Direct and indirect predictors of traumatic stress and distress in orphaned survivors of the 1994 Rwandan Tutsi genocide

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content
DIRECT AND INDIRECT PREDICTORS OF TRAUMATIC STRESS AND DISTRESS IN
ORPHANED SURVIVORS OF THE 1994 RWANDAN TUTSI GENOCIDE

by

Lauren Christina Ng

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PSYCHOLOGY)

August 2012

Copyright 2012 Lauren Christina Ng

ii
Table of Contents
List of Tables .................................................................................................................... iii
List of Figures ................................................................................................................... iv
Abstract ............................................................................................................................. v

Chapter 1: Linguistic Components of Genocide Testimonies Predict Trauma Symptoms

Chapter 1 Abstract ........................................................................................................... 1
Chapter 1 Introduction ..................................................................................................... 3
Chapter 1 Method ........................................................................................................... 12
Chapter 1 Results ............................................................................................................ 28
Chapter 1 Discussion ....................................................................................................... 36

Chapter 2: Risk Pathways from 1994 Rwandan Tutsi Genocide Exposure to Distress and
Traumatic Stress in Orphaned Heads of Household

Chapter 2 Abstract .......................................................................................................... 52
Chapter 2 Introduction ................................................................................................... 54
Chapter 2 Method ........................................................................................................... 63
Chapter 2 Results ............................................................................................................ 76
Chapter 2 Discussion ....................................................................................................... 90

References .................................................................................................................... 108

Appendix A: Reliability Coefficients for Coded 2002 Data ........................................... 116
Appendix B: Descriptive Information for LIWC Words ................................................. 118
Appendix C: IES-R Subscales and Descriptives .............................................................. 119
Appendix D: Chapter 1 Attrition Analyses .................................................................... 120
Appendix E: Chapter 1 Auxiliary Variables and Correlations with Outcomes .............. 121
Appendix F: Descriptives of Experienced Violence, Witnessed Violence, and
Witnessing Family Violence Subscales ..................................................... 122
Appendix G: Reliability Coefficients for Coded 2008/2009 Data ................................. 124
Appendix H: Scale Descriptives .................................................................................... 125
Appendix I: Chapter 2 Attrition Analyses ...................................................................... 129
Appendix J: Chapter 2 Auxiliary Variables and Correlations with Outcomes .............. 130

iii
List of Tables
Table 1. Correlations between Chapter 1 Study Variables ............................................. 24
Table 2. Descriptives of Chapter 1 Study Variables ....................................................... 28
Table 3. Significant Indirect Pathways in Model B (Chapter 1) ...................................... 32
Table 4. Descriptives of Chapter 2 Study Variables ........................................................ 77
Table 5. Correlations between Chapter 2 Study Variables. ........................................... 78
Table 6. Significant Indirect Pathways in Model B (Chapter 2). ..................................... 84
Table 7. Significant Indirect Pathways in Model C (Chapter 2). ..................................... 88

iv
List of Figures
Figure 1. Chapter 1 hypothesized mediation model ...................................................... 11
Figure 2. Results of Chapter 1 hypothesized model (Model A) ...................................... 30
Figure 3. Results of Chapter 1 exploratory model (Model B) ......................................... 31
Figure 4. Chapter 2 hypothesized model ........................................................................ 62
Figure 5. Results of Chapter 2 hypothesized model (Model A) ..................................... 82
Figure 6. Results of Chapter 2 final model (Model B) ..................................................... 83
Figure 7. Results of Chapter 2 posthoc model (Model C) ............................................... 87

v
Abstract
Millions of children grow into adulthood having experienced severe war and
ethnic conflict as children. One such group is orphaned child and adolescent survivors
of the 1994 Rwandan Tutsi Genocide, in which one-seventh of the Rwandan population
was murdered over the course of 100 days. After the genocide, many of these children
took on the responsibility of caring and providing for other child survivors. Research has
documented that child survivors of the genocide are at increased risk of mental health
concerns (Dyregrov, Gupta, Gjestad, & Mukanoheli, 2000; Schaal & Elbert, 2006).
However, differences in PTSD symptoms in Rwandan orphan survivors have not been
fully explained by genocide exposure.
This dissertation consists of two manuscripts that used path analysis to identify
modifiable factors that contribute to mental health outcomes for orphaned heads of
household (OHH) in Rwanda, over and above genocide exposure. Participants were 100
OHH who were members of a community organization. Data came from genocide
testimonies that were given in 2002 and assessments of post-genocide mental health
and risk factors that were collected in a 2008/2009 follow-up of 61 of the 100 original
participants. The first manuscript used cognitive models of posttraumatic stress
disorder (PTSD) as a framework for investigating whether linguistic components of
genocide testimonies predict PTSD symptoms. Results were somewhat consistent with
cognitive models and indicated that the way survivors described their genocide
experiences predicted PTSD symptoms six years later. The second manuscript
vi
investigated whether post-genocide social-ecological risk factors predict distress and
traumatic stress. Lack of education substantially predicted both distress and traumatic
stress, lack of resources significantly predicted lower educational attainment, and social
support predicted distress. After accounting for post-genocide risk factors, genocide
experiences still directly predicted distress and traumatic stress. Implications for
intervention and research are discussed.

1
Chapter 1: Linguistic Components of Genocide Testimonies Predict Trauma Symptoms
Chapter 1 Abstract
One-seventh of the Rwandan population was brutally butchered in the 1994
Rwandan Tutsi Genocide. In the years following the genocide, many survivors gave their
genocide testimonies as a means of bearing witness to the realities of their suffering
and survival. Although testimonies are often seen as a way to help society, it is possible
that their content could be used to help identify survivors who are at the highest risk for
PTSD symptoms. This study used path analysis to analyze whether the linguistic
components genocide testimonies mediated the association between genocide
experiences and PTSD symptoms 14 years after the genocide. Participants were 100
orphaned heads of household who were members of a Rwandan community
organization. Genocide testimonies were given in 2002 and PTSD symptoms were
assessed in a 2008/2009 follow-up of 61 of the 100 original participants. Linguistic
components of the genocide testimonies were measured using the Linguistic Inquiry and
Word Count program and genocide experiences were coded from testimonies. Results
indicated that after accounting for genocide experiences, somatosensory and
perceptual words negatively predicted intrusion symptoms and word count positively
predicted intrusion and hyperarousal symptoms. Moreover, body state words and word
count mediated the association between experienced violence and PTSD symptoms.
Results suggest that the way survivors described their genocide experiences contributed
just as much, if not more, to PTSD symptoms than the genocide exposure itself. Even
2
among survivors who have suffered some of the most severe genocide experiences,
linguistic analysis of genocide testimonies may help identify survivors at the highest risk
of developing PTSD symptoms.

3
Chapter 1 Introduction
Between 1901 and 1987, approximately 151 million people were killed in
genocides and mass murders, a number that dwarfs the 38.5 million people who were
killed in battle in all of the world’s international and civil wars during the same period
(Rummell, 1994). Soon after the report of those chilling statistics, another genocide
broke out in Rwanda. On April 6, 1994, after decades of increasing ethnic tension and
violence between the Hutu majority and Tutsi minority, a government-backed campaign
to exterminate the Tutsi population spread throughout Rwanda. Over the next 100
days, one-seventh of the population (approximately 1 million Tutsis and 50,000
moderate or sympathetic Hutus) was brutally butchered (Kigali Memorial Centre, 2012).
Survivors were exposed to extreme levels of physical and psychological violence
including rape, torture, mutilation, and witnessing their family members and loved ones
being brutally attacked and murdered (Human Rights Watch, 1999). As a result of these
experiences, traumatic stress symptoms related to the genocide were highly prevalent
in survivors (Dyregrov, Gupta, Gjestand & Mukanoheli, 2000; Pham, Weinstein &
Longman, 2004; Schaal & Elbert, 2006), and have been found to still be high years later
(Schaal & Elbert).
Testimony and Trauma Narratives
In the years following the 1994 Rwandan Tutsi Genocide, many survivors gave
their genocide testimonies, which are narrative records of their genocide experiences.
While all survivors have their own reasons for providing testimony, survivors of
4
genocides are often called upon to give testimonies as a means of bearing witness to
the realities of mass violence, with the goal of education, advocacy, and justice (Weine,
Kulenovic, Pavkovic & Gibbons, 1998). It is thought that individuals who give testimony
do so with the understanding, and often the hope, that their memories will become part
of collective knowledge and contribute to society’s acknowledgement of their suffering
and survival (Weine et al.). Although testimonies are often seen primarily as a way to
help society, it is possible that their content could be used to help the people giving the
testimonies.
Research on trauma-focused treatments has found that interventions that
involve asking individuals with posttraumatic stress disorder (PTSD) to describe
traumatic incidents from their past in narrative form are effective at reducing PTSD
symptoms (Bisson & Andrew, 2007; Van Etten & Taylor, 1998). In addition, participants
in studies who were asked to write about distressing experiences, have been found to
have better physical (and sometimes mental) health outcomes months later compared
to participants who wrote about neutral events (see Frattaroli, 2006 for a review).
These interventions and research studies typically used “narrative reliving” approaches
to help elicit the traumatic memories, which asks participants to describe the
traumatizing events as vividly and with as much detail as possible, including how they
felt, what they thought, what they saw, and everything about their surroundings that
they could remember, as if it were happening in the present (O’ Kearney & Perrott,
5
2006; Foa, Molnar & Cashman, 1995; Jones, Harvey & Brewin, 2007). These prompts
may influence the way that participants construct their trauma narratives.
In contrast to trauma narratives, individuals who give genocide testimony (not as
part of a legal trial) are typically not encouraged to describe their experience in
particular ways. Instead it is thought that survivors describe their genocide experiences
in more naturalistic ways, highlighting details that they choose to share and believe are
important for the world to know, with the primary purpose of documentation rather
than solace (Weine et al., 1998). Despite these differences, trauma narratives and
genocide testimonies are both detailed descriptions of distressing past events, and it is
possible that much can be learned about genocide survivors’ psychological states and
mental health through analysis of their genocide testimonies. Specifically, analyzing the
way that survivors construct testimonies may provide information about how survivors
are processing their traumatic experiences, which may predict their traumatic stress
symptoms years later. Therefore the words used in genocide testimonies might be used
to identify people who are at higher risk of mental health difficulties. This knowledge
could be particularly beneficial in situations of high mental health need and relatively
low mental health treatment resources, such as in post-genocide Rwanda.
Cognitive models of posttraumatic stress disorder
Cognitive models of PTSD suggest that trauma narratives may provide clues
about the way people process distressing events and make sense out of senseless
situations (Pennebaker & Seagal, 1999; Siegel, 1995). One such model (Ehlers and Clark,
6
2000) states that PTSD persists because of negative appraisals of trauma and its
sequelae. The model suggests that memories of distressing events have poor
contextualization and elaboration, strong perceptual priming, and strong associations,
which can lead to intrusive thoughts and persistent PTSD (Ehlers & Clark, Siegal). The
theory further predicts that negative appraisals result in strong and persistent negative
emotions (e.g. anxiety, depression or anger; Ehlers & Clark). In addition, disrupted
autobiographical memory leads to difficulty retrieving complete memories, thus
resulting in disjointed and poorly detailed accounts of traumatic events (Foa & Riggs,
1993; van der Kolk & Fisler, 1995; Amir, Stafford, Freshman & Foa, 1998). The inability
to remember details of a traumatic event is thought to maintain the sense of threat and
can lead to inaccurate and erroneous appraisals that exacerbate negative emotions
(Ehlers & Clark). Although individuals with PTSD may have difficulty intentionally
accessing memories of the traumatic event, they often experience involuntary intrusive
memories that are very emotional and vivid, and consist of sensory impressions rather
than thoughts (Ehlers & Clark).
Linguistic components of trauma narratives predicting PTSD symptoms
As hypothesized by cognitive theories of PTSD (Hellawell & Brewin, 2004;
Brewin, Dalgleish & Joseph, 1996; O’Kearney & Perrott, 2006), greater use of the
linguistic components of trauma narratives that represent somatosensory (i.e., body
states and symptoms), perceptual (i.e., touch, smell, hear), and negative emotion should
predict PTSD symptoms. Using a text analysis program, Linguistic Inquiry and Word
7
Count (LIWC: Pennebaker, Chung, Ireland, Gonzales, and Booth, 2007), researchers have
found support for these predictions. For example, Eid and colleagues (2005) found that
negative emotional expression in the trauma narratives of 120 soldiers involved in
military accidents was associated with greater PTSD symptoms. A second study of 28
female assault victims who developed trauma narratives as part of a PTSD intervention
found that death words in trauma narratives significantly predicted posttreatment PTSD
symptoms (Alvarez-Conrad et al., 2001). A third study of 104 non-clinical adults who
were asked to describe their worst traumatic experience found that total word count
was associated with less distress, and use of more body state words was associated with
greater distress and more PTSD symptoms (Beaudreau, 2007). Finally, results of a study
analyzing the trauma narratives of 131 road traffic accident survivors found that
survivors with PTSD had narratives with more sensory words than the narratives of
survivors without PTSD (Jones et al., 2007). Overall the results of the literature seem to
support the hypothesis that individuals who produce trauma narratives with more
negative emotional expression, death words, body states words, and more words
overall are more likely to report PTSD symptoms.
Although cognitive models of PTSD suggest that intrusion and reexperiencing
should be the aspects of PTSD most strongly predicted by linguistic components of
trauma narratives, only one study investigated the association between the three
components of PTSD (intrusion, avoidance, and hyperarousal) and linguistic components
of the narratives (Eid et al., 2005). While the Eid et al. study improved upon other
8
studies by investigating the three aspects of PTSD as separate outcomes, it did so using
Pearson correlation, which did not allow for the independent prediction of intrusion
after accounting for avoidance and hyperarousal.
Characteristics of distressing experiences
Findings indicate that trauma narratives may be amenable to linguistic analysis
and may be associated with PTSD symptoms. However, while some of the studies
include the type of trauma exposure as a variable (i.e., assault versus motor vehicle
accident), only one study included actual characteristics of the traumatic events as
predictors (Alvarez-Conrad et al, 2001). By not including characteristics of the traumatic
events, studies may be missing information that is more predictive of PTSD symptoms
than the linguistic components of the trauma narrative, particularly since most studies
included trauma narratives describing distressing events with very different
characteristics and time intervals since the trauma. Indeed, only two studies compared
the narratives of participants who were responding to the “same” traumatic experience
(Dekel & Bonanno: World Trade Center attack; Eid et al, 2005: military training exercise
accident). It may be that the primary predictor of PTSD symptoms is the specific
distressing experience (such as being injured or witnessing someone being attacked)
rather than the words used to describe those experiences. It is also possible that the
words in the trauma narratives may mediate or partially mediate the association
between characteristics of the traumatic experience and PTSD symptoms.
9
In addition to being the primary predictor of PTSD symptoms, the traumatic
events may also differentially predict different types of words used in genocide
testimonies. For example, survivors who are describing witnessing someone being
stabbed versus describing themselves being stabbed may use fewer emotion, sensory,
body state, and perceptual words, since the purpose of the testimony is to document
atrocities rather than to describe their own internal and personal experience. Similarly,
survivors who witnessed loved ones being harmed may report more sadness or anger
than participants who witnessed strangers being attacked.
Present study
This study analyzed the words used in genocide testimonies, and whether the
type of words mediated the association between 1994 genocide experiences and PTSD
symptoms in 2008/2009, 14 years after the genocide. Testimonies were given in 2002,
eight years after the genocide. The tested model hypothesized that words used in
genocide narratives would partially mediate the association between genocide exposure
and PTSD symptoms. Multiple types of genocide experiences were included to
determine whether they differentially predicted linguistic characteristics of trauma
narratives and PTSD symptoms.
The study hypothesized that genocide experiences would positively predict
negative emotion, somatosensory, perceptual, death words, and word count, and in
turn greater use of somatosensory and perceptual words would positively predict
intrusion symptoms, while greater use of negative emotion, death words, and words
10
overall would positively predict intrusion, avoidance, and hyperarousal. Somatosensory
and perceptual words were predicted to correlate with each other, and negative
emotion words were predicted to correlate with each other. The three genocide
exposure variables were hypothesized to correlate together, as were the three PTSD
outcomes.
Finally, the model also includes sex as a predictor of negative emotion, as
research has shown that women are more likely to engage in more verbal emotion-
focused coping than men (Tamres, Janicki, & Helgeson, 2002; Ptacek, Smith, & Zanas,
2006), and might be expected to use more negative emotion words when describing a
traumatic event. Age was included as a predictor of somatosensory and perceptual
words, because while research has found that children as young as 3 years old can give
reasonably coherent accounts of past memories (Fivush, 1998), research has also found
that people who experienced traumatic events at younger ages had a harder time
remembering and recalling the details of those experiences (Williams, 1994). The
hypothesized model is pictured in Figure 1.

11
Figure 1. Chapter 1 hypothesized mediation model.

Note. All paths between genocide exposure variables and LIWC variables are hypothesized to be positive, and all
paths between LIWC variables and PTSD symptoms are hypothesized to be positive.

12
Chapter 1 Method
Participants
Participants were 100 orphaned heads of household (OHH) who were members
of the Rwandan Association des Orphelins Chefs de Ménages (AOCM) (i.e. the
Association of Orphans Chiefs of Household). Sixty-one of them also participated in the
2008/2009 follow-up. 58% were male and in 2002 they ranged in age from 13 to 35,
with a mean age of 22.25. Therefore, during the genocide they ranged in age from 5 to
27, with a mean age of 14.25. In 2002 they were caring for 2.23 children on average. By
2008/2009, 35% of the participants were married and 25% had biological offspring.
Procedures
The data from this study come from two sources. The first is the AOCM
Genocide Oral History Project which was conducted in 2002 by AOCM, in collaboration
with Dr. Donald Miller of the USC School of Religion and Ms. Lorna Miller of All Saints
Episcopal Church in Pasadena, CA. The second is the AOCM-USC Trauma Project which
was conducted in 2008/2009 by AOCM, Dr. Beth Meyerowitz of the USC Department of
Psychology, and Dr. Miller (Meyerowitz et al., 2010). The AOCM-USC Trauma Project
followed up on the participants in the 2002 AOCM Genocide Oral History Project.
In 2002, approximately 8 years after the genocide, Dr. Miller and AOCM agreed
to collaborate on an oral history project of AOCM members, resulting in the AOCM
Genocide Oral History Project. Dr. and Ms. Miller spent a week with the AOCM
leadership developing an interview guide and having them practice interviewing
13
techniques and strategies. The interview protocol was reviewed and approved by an
umbrella organization of genocide survivor associations of which AOCM was a part.
Over the next six months, AOCM researchers used semi-structured interviews to collect
and record 100 genocide testimonies of AOCM members across Rwanda. The
participants were selected by the AOCM board from their list of beneficiaries to
represent a wide range of their members, with equal numbers of males and females,
from each of Rwanda’s provinces in both rural and urban environments. Participants
were approached and asked if they wanted to give testimony of their genocide
experiences.
The AOCM Genocide Oral History Project interviews were conducted in
participants’ homes, in locations in the villages near their homes that were selected by
the participants, or in the AOCM headquarters in Kigali City. In addition to genocide
testimonies, interviewers recorded participants’ self-reported demographic information.
Genocide testimonies and interviews were conducted in Kinyarwanda by other
members of AOCM and were audiorecorded. The audiorecordings were translated and
transcribed into English by one native Kinyarwanda speaker. Three of the genocide
testimonies could not be located, and so 2002 information for these three participants is
limited to demographic information.
From August 2008 to May 2009 (approximately 14 years after the genocide), the
AOCM-USC Trauma Project attempted to collect data on mental health problems and
risk factors from 99 of the original 100 Genocide Oral History participants (one of the
14
2002 participants asked to remain anonymous and was therefore not included in the
follow-up). Interviewers sought out participants at their last known village or address.
If the original participants could not be located, interviewers asked neighbors or friends
for information about their possible whereabouts. Once located, the purpose and
procedures of the follow-up study were explained and participants gave informed
consent. All study methods were approved by the University of Southern California
Institutional Review Board and the AOCM board.
For the 2008/2009 AOCM-USC Trauma Project, interview questions and
measures were selected from existing validated measures and were developed by the
USC researchers in collaboration with the AOCM board. To assist with measuring
constructs in culturally appropriate ways, focus groups of OHH were convened to gather
information about their day-to-day lives, experiences, and mental health concerns. In
addition, participants in the focus groups also evaluated the face validity of standard
measures and assessed their cultural appropriateness. The focus group responses
assisted with preliminary development of scale items and interview questions, which
were then revised and approved by the AOCM board members. After approval by the
board, all measures and interview protocols were forward- and back-translated
between English and Kinyarwanda by native Kinyarwanda speakers to ensure accuracy.
After the data were collected, the translations of the measures and interview questions
were reviewed by a second set of translators and if a consensus was not reached on the
accurate translation, then the items were dropped. Six AOCM members were selected
15
as interviewers by the AOCM board, and participated in a week-long training on how to
conduct semi-structured interviewers by Dr. Meyerowitz and the AOCM project
director, Mr. Naphtal Ahishakiye.
Interviews were conducted in Kinyarwanda. Interviews were audio recorded and
interviewers wrote participants’ verbal responses to instrument questions directly on
the questionnaires. Forty-five of the 61 audio recorded Kinyarwanda interviews were
transcribed in Kinyarwanda, and then a native Kinyarwanda speaker (selected by AOCM)
who was also fluent in English recorded the interviews orally in English. The English
audio recordings were then transcribed by USC undergraduates who were native English
speakers. The accuracy of English transcripts was reviewed by a second native English
speaker. Two other native Kinyarwanda speaking translators were hired by USC to
complete the remaining translations, and they wrote English transcripts using the
original Kinyarwanda audio recordings. Eleven of the interviews were translated by two
different translators, which allowed for qualitative comparison between translators.
While exact wording of the translations differed somewhat, the content was consistent.
Sixty-three of the original participants were located and 61 agreed to participate
in the follow-up study. Of the 36 people from the 2002 sample who could not be
located for the follow-up, information was available about the presumed whereabouts
of 12: one had joined the military and 2 were away at jobs and school in other
countries, seven had relocated and could not be found, and two were in psychiatric
16
hospitals. No one could be located to provide information on the remaining 24
participants.
Interview Coding
Four female undergraduate research assistants, who were blind to study objectives and
outcomes, coded the 2002 genocide testimonies in English. In order to efficiently code
the transcripts, sequential overlapping coding for reliability testing was used
(Neuendorf, 2001). Therefore, every interview was coded by two coders, with six
possible combinations of coders. Research assistants were trained on how to interpret
and score each variable. In cases where discrepancies or missing information remained
even after each coder independently reviewed her own coding, items were reviewed by
the author who made the final decision based on a close reading of the interview. Out
of 4,365 codes (97 genocide testimonies and 45 variables), the author made a final
decision on 175 (4%). The 2002 interviews were semi-structured, with only limited
prompting (i.e., “Tell me about your genocide experience?”), and so items were coded
as “Yes” or “Not mentioned”, and there were no missing data recorded. All coded items
were dichotomous. Interrater reliability was acceptable for all items (Cohen’s Kappa >=
0.50, ICC >= 0.60; Stemler & Tsai, 2008). Cohen’s Kappa for dichotomous items and
Intraclass Correlation Coefficients (ICC) for individual items are reported in Appendix A.
Study Measures
Sociodemographics
Participants’ Sex and Age were recorded by the 2002 interviewers.
17
Genocide Experiences
Genocide experiences were coded from the 2002 genocide testimonies. Coding
sheets were developed from the Rwandese Children’s Exposure to War Scenes (RCEWS)
Measure (Dyregrov et al., 2000), which is a 30-item measure that assesses exposure to
events that were common during the Rwanda Genocide (i.e. “Did you witness people
being massacred?). In order to ensure that the coding sheets were comprehensive,
specific to OHH in this study, and included genocide experiences regardless of their
frequency and association with other genocide events (Netland, 2001; Netland, 2005;
Layne et al., 2010), additional items were added to the coding sheets that were relevant
to the AOCM sample (e.g., witnessing people being thrown into a pit or latrine;
witnessing people being drowned; witnessing their mothers being attacked). Finally, to
assess experienced violence, parallel items were included regarding the participants’
direct experiences of each genocide event.
The coded items were divided theoretically into the sum-scale composites (1)
Witnessed Violence and (2) Experienced Violence. The Experienced Violence subscale
was not completely parallel to the Witnessed Violence subscale, because some items on
the Witnessed Violence subscale are impossible to have experienced and survived (e.g.,
witnessing someone being killed). Additionally, one Experienced Violence item (being
raped with an object) was dropped because no participants endorsed it. The Witnessed
Violence subscale was composed of 22 genocide exposure items and the Experienced
Violence subscale was composed of 19 items.
18
A third subscale was created (3) Witnessed Harm to Family, that assessed
whether or not participants mentioned witnessing different immediate family members
being attacked. Due to the nature of the descriptions of the genocide narratives, it was
often unclear whether participants were describing the same sibling or family member
being attacked or a different family member, and so coding was limited to four items:
mentioning witnessing an attack on your 1) mother, 2) father, 3) sibling, or 4) other
relative, and so the range in this subscale is limited to 0 to 4 and probably
underestimates the number of immediate family members whom participants
witnessed being attacked. Reliability for all scales was good. ICCs for Witnessed
Violence, Experienced Violence, and Witnessed Harm to Family were .93, .92, and .92,
respectively.
Linguistic Components of Genocide Testimonies
In order to quantify how participants talked about their genocide experiences, a
text analysis program, Linguistic Inquiry and Word Count (LIWC: Pennebaker et al.,
2007), was run on the portion of the 2002 interview transcripts in which participants
described their genocide experiences (the first section of the interview asked about
details of the participant’s family and history). Interviewer statements were deleted
from the transcripts. Transcribed genocide experiences were cleaned according to the
suggestions of the LIWC manual, and then run through the 2007 version of LIWC.
LIWC2007 is a content-analysis computer program composed of a text-
processing mechanism and a support dictionary (Pennebaker et al., 2007). The support
19
dictionary of LIWC2007 is composed of approximately 4,500 words or word stems
assigned to subdictionaries. The subdictionaries are composed of words grouped to
reflect a particular aspect of language. To validate the variables produced by the
program, Pennebaker and colleagues had judges assign words to linguistic categories
and then the assignments were compared. Judge agreement of category assignment
was good (Pennebaker et al.). With the exception of total word count, for each piece of
text, LIWC computes the percentage of total words that are represented by each
linguistic category (Pennebaker, et al.). For example, if a participant used 10 words that
were categorized as Anger words in a 500 word narrative, then that participant’s LIWC
anger score would be 2 (or 2%).
The hypothesized model used nine linguistic categories that represented
perceptual, sensory, negative emotion words, or have been found to be associated with
PTSD symptoms in the literature. The nine categories are 1) seeing, 2) hearing, 3)
touching
1
, 4) body states, 5) anxiety, 6) anger, 7) sadness, 8) death words, and 9) word
count. As each of these categories represents a very small number of possible words
participants could use, their percentages are expected to be small. See Appendix B for
the number of words in the LIWC dictionary for each category and Cronbach’s alphas
calculated by Pennebaker et al. (2007).

1
In the LIWC 2007 Dictionary the variable “Touching” is called “Feeling” (Pennebaker et al., 2007)
20
PTSD Symptoms
PTSD symptoms were assessed in 2008/2009 using the Impact of Events Scale-
Revised (IES-R: Weiss & Marmar, 1997). The IES-R is a twenty-two item self-report
questionnaire that measures traumatic stress. Although the IES-R was not developed
for use in a Rwandan sample, it has been used in post-genocide Rwandan samples
(Dyregrov et al., 2000). The IES-R asked participants to rate the distress of each item
“with respect to the genocide and your current living situation” over the past two
months. Items are scored from 0 (“not at all”) to 4 (“extremely”). In addition to a total
score, the 22-items can be divided to create subscale scores for each of the three
aspects of PTSD: intrusion, hyperarousal, and avoidance symptoms. The subscale scores
are the means of their respective items.
Intrusion
Intrusion was the mean of seven of the eight IES-R items that compose the
intrusion subscale (i.e., “Any reminder brought back feelings about the genocide”). Item
2 (“I had trouble staying asleep”) was dropped from the scale because in Kinyarwanda it
could not be distinguished from item 15 (“I had trouble falling asleep”) which is on the
hyperarousal scale. Item 15 was also not included on the hyperarousal scale. The
results of a reliability test indicate that the Cronbach alpha score for the 7-item intrusion
scale was .87.

21
Hyperarousal
Hyperarousal was the mean of five of the six IES-R items that compose the
hyperarousal subscale (i.e., “I was jumpy and easily distracted”). As noted above, item
15 was dropped from the scale since it could not be distinguished from an item that
typically loads on the intrusion subscale. The results of a reliability test indicate that the
Cronbach alpha score for the 5-item hyperarousal scale was .78.
Avoidance.
Avoidance was the mean of seven of the eight IES-R items that compose the
avoidance subscale (i.e., “I stayed away from reminders of the genocide”). There was
disagreement between translators over the translation of Item 17, which was intended
to ask whether participants endorsed “I tried to remove the genocide from my
memory,” and so this item was dropped from the scale. The results of a reliability test
indicate that the Cronbach alpha score for the 7-item avoidance scale was .61. See
Appendix C for IES-R items and their descriptives.
Preliminary Analyses
Attrition Analyses
To assess whether any 2002 variables predicted attrition, logistic analyses were
run in which all 2002 variables were included as predictors of 2008/2009 completion.
Using an alpha of p<.10, logistic regression comparisons between subjects retained and
those lost to follow-up revealed that Experienced Violence was the only 2002 variable
that significantly predicted follow-up completion, such that those who experienced
22
more violence were more likely to complete the follow-up interview (p=0.07, mean
Experienced Violence for follow-up completers = 4.29 and for follow-up non-completers
= 3.55). No other 2002 variables significantly predicted follow-up completion. See
Appendix D for attrition analysis results.
Normality of study variables
Study variables were screened for outliers, skewness, and kurtosis. Witnessed
Harm to Family, Word Count, Sadness, Anxiety, Body, and Touching did not meet the
assumption of normality and were transformed. Word Count was successfully
transformed using the logarithmic transformation, and Sadness and Body were
successfully transformed using the inverse transformation. The transformed variables
will be used in all future analyses. Witnessed Harm to Family, Anxiety, and Touching
remained non-normal despite use of all transformations. These remaining three
variables were used in their original form, with the limitation that they are non-normal.
Auxiliary Variables
In order to reduce bias and increase power, auxiliary variables (covariates that
help predict missing values but are not in the tested model) were included in all
analyses (Graham, 2009). To ensure that the model was predicting missing variables as
accurately as possible, variables in the broader data set that were not used in this study
but were significantly associated with dependent variables at r≥0.30 were considered
auxiliary variables (Enders, 2008). Bivariate correlations between the predicted
variables used in this study (LIWC variables, intrusion, hyperarousal, and avoidance) and
23
variables in the larger dataset that were not used in this study were run to identify
auxiliary variables that met the criteria outlined above. This process identified seven
variables that met the above criteria for at least one of the variables in the study. The
seven auxiliary variables were 1) family deaths (the number of immediate family killed),
2) percent of immediate family killed, 3) distress in 2008/2009, 4) lack of resources in
2008/2009, 5) lack of education in 2008/2009, and two LIWC words that were
associated with predicted variables but were not hypothesized in the model: 6) health
words, and 7) sexual words. See Appendix E for list of auxiliary variables and their
correlations with variables used in this study.
Latent Variables Underlying Linguistic Components of Genocide Testimonies
The original model hypothesized that the somatosensory words (body, seeing,
hearing, and touching) would predict an underlying latent variable representing
somatosensory experiences, and the negative emotion words (anger, sadness, and
anxiety) would predict an underlying latent variable of negative emotion. However, the
correlations between these variables were not significant, and therefore confirmatory
factor analyses that attempted to fit these latent variables (both together and in
separate models) did not converge (see Table 1 for correlations between study
variables).

Table 1. Correlations between Chapter 1 Study Variables
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1. Male
--- .07 -.18 -.10 -.09 -.09 .15 -.08 .11 -.10 -.09 -.02 .01 -.16 -.12
2. Age
--- -.08 -.29
**
-.01 .00 -.15 .02 .21
*
-.14 -.15 -.02 .04 -.10 .13
3. Witnessed Violence
--- .34
***
.44
***
.33
***
-.08 .24
*
-.05 -.04 .15 .11 -.33
***
.23
*
.28*
4. Witnessed Harm to Family

--- .23
*
.18 .03 .05 -.06 .00 -.08 .07 -.22
*
.11 -.11
5. Experienced Violence

--- .43
***
-.12 -.06 -.15 -.09 .05 .10 -.53
***
-.07 .16
6. Word Count

--- -.02 -.28
**
.07 -.18 .20
*
.01 -.26
**
-.37
***
.30*
7. Anxiety

--- -.03 -.01 .08 .05 -.04 .12 -.09 -.01
8. Anger

--- -.14 .03 -.11 -.07 .12 .76
***
.16
9. Sadness

--- .03 -.04 -.01 .04 -.19 .13
10. Seeing

--- -.02 .13 -.03 .00 -.16
11. Hearing

--- -.14 -.03 -.12 -.03
12. Touching

--- -.28
**
-.11 .21
13. Body

--- .07 -.33*
14. Death

--- .09
15. Intrusion

---
16. Avoidance

17. Hyperarousal

*p<.05, **p<.01, ***p<.001

24

Table 1 continued
16 17
1. Male
-.23 -.07
2. Age
-.01 .00
3. Witnessed Violence
.26* .43**
4. Witnessed Harm to Family
.05 .04
5. Experienced Violence
.13 .28*
6. Word Count
.11 .37**
7. Anxiety
-.14 .08
8. Anger
.15 .12
9. Sadness
.10 .17
10. Seeing
.04 -.14
11. Hearing
.07 .28*
12. Touching
.36** .21
13. Body
-.30* -.31*
14. Death
.18 .09
15. Intrusion
.49*** .71***
16. Avoidance
--- .66***
17. Hyperarousal
---
*p<.05, **p<.01, ***p<.001

25
26
Exploratory factor analyses using all nine linguistic components found that a
model with three latent variables fit the data well χ
2
(12)=3.95, p=.98, CFI=1.00, TLI=1.24,
RMSEA=.00, SRMR=.02, however, for the first factor, only two variables had rotated
factor loadings >.60 (Anger=.61 and Death =1.14), and for the second factor, only one
item had a factor loading >.60 (Body =.79). The other remaining items had low primary
factor loadings (i.e. less than .35), or loaded onto two factors equally. It was therefore
concluded that the linguistic components variables did not hang together well enough
to represent underlying latent categories, and were therefore used as independent
variables.
Analysis Plan
Path analysis was used to examine associations between the variables and
consider the mediating role of the LIWC variables in the hypothesized model (see Figure
1). Models were run using SEM with Mplus statistical modeling software (Version 6.12;
Muthén & Muthén, 2011). SEM allows for all of the associations of the variables to be
tested simultaneously, direct and indirect effects to be identified, the fit of the overall
model to be tested, and the use of auxiliary variables to reduce bias and increase power.
In order to correct for some non-normality found in some variables, the analyses used
full information maximum likelihood (FIML) estimation with robust standard errors
(MLR estimator). Even for small sample sizes, FIML has been found to perform
reasonably well (Hoyle & Panter, 1995). In addition, FIML handles missing data,
assuming data are missing at random.
27
In order to assess the fit of the hypothesized models, the structural path model
shown in Figure 1 was tested wherein all hypothesized paths shown were estimated
freely and all other possible paths not shown were fixed at 0 (Model A). Goodness of fit
was determined using the comparative fit index (CFI), Tucker-Lewis index (TLI), root-
mean-square error of approximation (RMSEA), and standardized root-mean-square
residual (SRMR). Indicators of acceptable model fit for small sample sizes are considered
to be a CFI > .95, TLI> .95, RMSEA < .06, and SRMR < .09 (Hu & Bentler, 1999).
After testing the hypothesized model, exploratory analyses were conducted to
determine if the fits of the models could be improved with alternative paths. The
exploratory analyses were assessed through a three-step procedure. First, a just-
identified model was tested with all potential paths in the hypothesized models left free
to vary. Second, model trimming was conducted on the just-identified models following
procedures recommended by Chou and Bentler (2002). Estimated regression
coefficients were inspected, paths with the smallest z-score were systematically fixed to
zero and the model was computed again, until all paths were significantly different from
zero (z>1.96, p<0.05). Once all regression paths were significant, the covariate matrix
was inspected, and non-significant correlations were also fixed to zero. The model
resulting from the exploratory analyses is Model B.

28
Chapter 1 Results
See Table 2 for descriptive of study variables.
Table 2. Descriptives of Chapter 1 Study Variables
Measure N
Mean or
% SD
Observed
Range
Possible
Range
2002 sociodemographics

Male 100 58.00%

Age 88 22.25 5.06 13 – 35

1994 Genocide experiences

Witnessed violence 97 7.09 3.05 0 – 14 0 - 22
Experienced violence 97 3.97 1.91 0 – 9 0 - 19
Witnessed harm to family 97 1.27 1.27 0 – 4 0 - 4
Words Used in 2002 Genocide Narrative

Seeing (View, saw, seen) 97 0.97 0.50 0 - 2.40
Hearing (Listen, heard) 97 0.55 0.31 0 - 1.40
Touching (Feels, touch) 97 0.14 0.15 0 - 1.00
Body (Cheek, hands, spit) 97 0.35 0.28 0 - 1.52

Anxiety (Worried, fearful, nervous) 97 0.14 0.15 0 - .78

Anger (Hate, kill, annoyed) 97 2.48 0.75 .84 - 4.49

Sadness (Crying, grief, sad) 97 0.24 0.20 0 - 1.25

Death (Bury, coffin, kill) 97 2.81 0.79 1.07 - 5.08

Word Count 97 870.12 402.97 350 – 2091

2008/2009 PTSD Symptoms

Intrusion 61 2.51 1.01 0 – 4 0 - 4
Avoidance 61 2.18 0.80 0 – 4 0 - 4
Hyperarousal 61 2.12 1.11 0 – 4 0 - 4
Note. Untransformed variables are presented for ease of interpretation.

Genocide Experiences
Participants witnessed an average of 7.09 of 22 different genocide events and
personally experienced an average of 3.97 of 19 different genocide events. The most
frequently endorsed Witnessed Violence items were witnessing someone being killed
(94%), witnessing someone being attacked (89%), seeing dead bodies (72%), and
witnessing massacres (many people killed at one time) (58%). The most frequently
29
endorsed Experienced Violence items were having your house damaged or destroyed
(95%), being threatened to be killed (84%), being attacked or assaulted (62%), and being
injured (41%). 30% of participants reported witnessed their mothers being attacked and
26% witnessed their fathers being attacked. 42% of participants witnessed their siblings
being attacked. 29% witnessed other family members being attacked. Overall, 63% of
participants reported witnessing at least one family member being attacked. See
Appendix F for list of genocide exposure items and their descriptives.
LIWC Variables
On average, participants spoke 870 words during their trauma narrative, and the
most prevalent word categories were words related to death (2.8% = approximately 24
words) and words expressing anger (2.5% = approximately 22 words).
PTSD Symptoms
The overall mean IES-R score was 2.29 out of 4, with 82% of participants having
scores at or above 1.5, which has been found to be the IES-R score with the best
diagnostic accuracy for assessing PTSD in Vietnam Veterans (Creamer, Bell & Failla,
2003). All of the subscale scores were highly elevated, with participants endorsing
experiencing intrusion, avoidance, and hyperarousal symptoms between “Somewhat”
and “A lot” on average. Intrusion symptoms were the most prevalent out of the three
subscales.

30
Model Testing
The results indicated that the hypothesized model did not fit the data,
χ
2
(52)=166.02, p<.001, CFI=.60, TLI=.03, RMSEA=.15, SRMR=.12, and there were several
paths that were non-significant. The results of the hypothesized structural path model
(Model A) are shown in Figure 2.
Figure 2. Results of Chapter 1 hypothesized model (Model A).

Note. All coefficients are standardized. Paths not shown were not significant.
*p<.05. **p<.01. ***p<.001.

31
Results of the exploratory analysis fitting the model using the model trimming
procedure (Model B) are shown in Figure 3.
Figure 3. Results of Chapter 1 exploratory model (Model B).

Note. All coefficients are standardized. Dashed lines represent statistically significant paths that do not
account for a significant amount of variance in the outcomes.
*p<.05. **p<.01. ***p<.001.

Results indicate that the exploratory model fit the data, and that all of the fit
indices except for SRMR were within acceptable limits, χ
2
(106)=115.03, p=.26, CFI=.97,
TLI=.96, RMSEA=.03, SRMR=12. Although the SRMR score is above the suggested cut-
32
offs (Hu & Bentler, 1999), unlike the other fit indices, SRMR is biased with respect to
sample size, such that smaller sample sizes result in inflated SRMR values and high
rejection rates (Marsh, Hau, & Wen, 2004). Given that the other fit indices are within
acceptable limits, it is likely that the model is still an acceptable fit to the data. Since the
exploratory model fit the data better and is more parsimonious than the hypothesized
model, it was considered the final model.
The final model (Model B) accounted for 26% of the variance in Intrusion, 12% of
the variance in Hyperarousal, 29% of the variance in Body States, 19% of the variance in
Word Count, 14% of the variance in Death words, and 13% of the variance in Anger.
There was no significant prediction of the variance in Avoidance, Seeing, Hearing,
Touching, Sadness, or Anxiety. Indirect effects of covariates and genocide experiences
on the PTSD subscales via the linguistic components of the genocide testimonies were
tested. Results of significant indirect pathways are presented in Table 3.
Table 3. Significant Indirect Pathways in Model B (Chapter 1)
Indirect pathways β SE B SE Z-score
Intrusion

Effect of experienced violence via word
count
0.14 0.05 1.46 0.55 2.66**
Effect of experienced violence via body
states
0.12 0.06 1.2 0.56 2.15*
Hyperarousal

Effect of experienced violence via word
count
0.11 0.05 1.13 0.55 2.07*
Note. Z-scores correspond to unstandardized Beta.

33
Results partially supported the hypothesis that the words used in the genocide
narratives would mediate the association between genocide exposure and PTSD
symptoms, as two of the LIWC variables, word count and body states, did fully mediate
the association between experienced violence and PTSD symptoms. In addition to the
indirect effect, two direct effects between genocide exposure and PTSD symptoms were
also significant: witnessed harm to family members negatively predicted intrusion
symptoms, and witnessed violence positively predicted hyperarousal symptoms. The
hypothesis that somatosensory words and perceptual words would only predict
intrusion symptoms was also partially supported. Body states, the only somatosensory
variable that was a significant mediator, only predicted intrusion symptoms. While
hearing was not significantly predicted by any genocide exposure variables, it did
significantly predict intrusion symptoms. The regression path from touching to
avoidance was significant, but touch did not predict a significant amount of the variance
in avoidance.
The hypothesis that word count would mediate the association between
genocide experiences and all three PTSD subscales was also partially supported. Word
count mediated the association between experienced violence and intrusion and
hyperarousal, but not avoidance. However, the hypothesis that negative emotion and
death words would mediate the association between genocide experiences and all three
outcomes was not supported. While witnessed violence and experienced violence

34
predicted the use of anger and death words, neither of these variables predicted any of
the PTSD outcomes.
The hypothesis that sex would predict negative emotion was not supported, as
sex was not associated with any of the variables in the model. The hypothesis that age
would predict somatosensory and perceptual words was also not supported. While the
regression path from age to sadness was significant, age did not predict a significant
amount of the variance in sadness. Finally, the hypotheses that the somatosensory
words would correlate with each other and the negative emotion words would correlate
with each other was not supported. However, word count, anger words, and death
words were all correlated, and anger and death words were very highly correlated,
sharing 57% of their variance.
Although the attrition analyses revealed that of the 2002 variables, only
Experienced Violence was associated with completion of the follow-up study, there may
have been differences in completers versus non-completers that were not measured in
this study. In order to assess the degree to which data from complete cases differed
from data from incomplete cases, the final model was run using only complete cases
and model fit was assessed. The model using only complete cases (N=61) fit the data
almost as well as the model with all cases (N=100): χ
2
(106)=115.67, p=.24, RMSEA=.04,
SRMR=.15. CFI and TLI could not be calculated because the ratio of observations to
estimated parameters was too low. In addition, the SRMR was further inflated due to
the substantial reduction in sample size. Despite the significant drop in power, the
35
model with the complete cases replicated most of the significant paths found in the final
model.
Four of the 12 paths that were significant in the final model were no longer
significant when the model was fit to complete cases: Experienced Violence predicting
Word Count (β=.24, p=.08), Experienced Violence predicting Anger (β=-.14, p=.29),
Experienced Violence predicting Death (β=-.17, p=.18), and Age predicting Sadness
(β=.15, p=.16), In the final model, Age did not predict a significant amount of variance in
Sadness, and so it is not surprising that in a model with reduced power, this path is no
longer significant. Moreover, all of the other paths were regressed on Experienced
Violence, which was the one variable that was identified as being predicative of
missingness. It was expected that its associations with other variables would change if
only complete cases were analyzed. Despite these limitations, the model still fit the
complete cases data well, which supports the model’s validity despite the large amount
of missing data. Moreover, since Experienced Violence was included in all analyses,
FIML is able to account for the differences in scores on Experienced Violence between
completers and non-completers when estimating the patterns of associations.

36
Chapter 1 Discussion
The participants in this study reported experiencing and witnessing many
horrifying atrocities during the 1994 Rwandan Tutsi Genocide, yet they also reported
having a wide range of PTSD symptoms 14 years later. This study sought to clarify why
some survivors have fewer PTSD symptoms than others by analyzing whether the
hypothesis put forth by cognitive models of PTSD, that the way participants described
their genocide experiences would predict PTSD symptoms, held true even in a sample
that has been exposed to such severe events. Cognitive models of PTSD suggest that
intrusion symptoms should be predicted by somatosensory and perceptual linguistic
components of trauma narratives, and negative emotion, word count, and death words
should predict intrusion, hyperarousal and avoidance (Ehlers & Clark, 2000; O’Kearney &
Perrott, 2006). In the final model, two somatosensory/perceptual categories (body
states and hearing) only predicted intrusion symptoms, and word count predicted both
intrusion and hyperarousal. Therefore the results partially supported the cognitive
models, as different types of linguistic components of genocide narratives predicted
different aspects of PTSD.
Linguistic components of genocide testimonies predicting PTSD symptoms
Body state words significantly predicted intrusion symptoms, and hearing words
significantly predicted hyperarousal symptoms, however the direction of the effects
were the opposite of those suggested by cognitive models of PTSD. Specifically, the use
of somatosensory and perceptual words negatively, rather than positively, predicted
37
intrusion symptoms. One possible explanation for this result is that the more
traumatized survivors are, the more difficulty they have retrieving complete detailed
memories of traumatizing events (Foa & Riggs, 1993; van der Kolk & Fisler, 1995; Amir et
al., 1998), which results in providing less somatosensory and perceptual information in
their genocide testimonies.
Another interpretation is that traumatized survivors are consciously or
unconsciously suppressing details of distressing events (Ehlers & Steil, 1995). However,
if this hypothesis were accurate, then a significant association between somatosensory
words and avoidance would be expected. However, none of the variables in the model
significantly predicted avoidance.
2
Notably, the internal reliability of the avoidance
subscale was moderate (Cronbach’s alpha=.61), and much lower than that of intrusion
(Cronbach’s alpha=.87) or hyperarousal (Cronbach’s alpha=.78). It may be that
avoidance is not as culturally-relevant in post-genocide Rwanda as intrusion or
hyperarousal. Indeed, it is very hard, if not impossible, to avoid reminders of the
genocide in Rwanda, as the entire country was impacted and many survivors still live in
or near the locations where they and their family were attacked, and often must co-
exist with people who were involved in the killings. Given that the social environment in
post-genocide Rwanda makes avoidance very difficult, and that the internal reliability
was not strong, caution must be used in interpreting the results with respect to
avoidance.

2
Touching did predict avoidance, but it did not account for a significant amount of its variance (6%, p=.3).
38
Contrary to hypotheses, negative emotion words did not predict PTSD
symptoms. Although anger was significantly predicted by witnessing and experiencing
violence, it did not go on to predict PTSD symptoms. It may be that for genocide
survivors, the erroneous cognitive appraisals thought to underlie negative emotion and
PTSD symptoms may be fairly accurate. For example, the negative appraisals associated
with “loss of trust in the others” are often associated with PTSD (Ali, Dunmore, Clark, &
Ehlers, 2002; Jaycox, Zoellner, & Foa, 2002). However, genocide survivors in Rwanda
continue to live in a world in which the government, neighbors, and millions of fellow
citizens attempted to destroy them, their families, and millions of others because of
their ethnicity. It may be that loss of trust in the others is not only reasonable, but it
may also be accurate. Indeed, the negative emotion that was most salient in this
sample was anger, which was correlated very highly with death words (r=.75).
Understandably, participants who described massive amounts of murder and death
were more likely to express anger. Therefore, rather than being an indicator of PTSD
symptoms, negative emotion in genocide testimonies may be a normative response to a
senseless situation. Moreover, it may be that negative emotion in genocide testimonies
would positively predict distress rather than PTSD symptoms.
As hypothesized, word count positively predicted intrusion and hyperarousal
symptoms, however word count did not predict avoidance symptoms. One
interpretation of this result is that traumatic events disrupt autobiographical memory
leading to difficulty remembering details of traumatic events and the development of
39
more fragmented, less coherent genocide narratives (Foa & Riggs, 1993; van der Kolk &
Fisler, 1995; Amir et al., 1998), which might be represented by longer narratives. In
turn, these fragmented memories maintain a sense of threat, resulting in PTSD
symptoms years later (Ehlers & Clark, 2000). Word count was positively predicted by
experienced violence, and although it could be argued that survivors used more words
because they had more genocide experiences to describe, the results indicated that
witnessing violence and witnessing harm to family members were not associated with
word count. Moreover, participants reported more incidents of witnessed violence
(mean=7) than experienced violence (mean=4), and therefore if higher word count was
solely a representation of more genocide experiences, then witnessed violence would
be expected to be positively associated with word count.
Characteristics of the genocide experience
Studies investigating the association between linguistic components of trauma
narratives and PTSD symptoms often fail to include trauma experiences as predictors.
Trauma experiences may differentially predict the linguistic components of genocide
testimonies, and thereby indirectly predict PTSD symptoms. Results indicated that
different types of genocide experiences predicted different types of linguistic
components of the genocide testimonies. As noted above, experienced violence
negatively predicted body state words and positively predicted word count. In addition,
experienced violence negatively predicted anger and death words.
40
It may be that survivors who personally experience many different forms of
violence (cutting, beatings, drowning, rapes, etc.) may be willing to report that those
events happened, but may consciously or unconsciously suppress the details of those
experiences, resulting in a negative association between experienced violence and body
states, anger, and death. These results are supported by the finding that the fewer body
state words used, the more intrusion symptoms survivors reported. In contrast,
witnessed violence positively predicted the use of anger and death words in genocide
testimonies, but anger and death words did not predict PTSD symptoms. Perhaps
witnessing violence does not result in the same degree of suppression of details as
experiencing violence, thereby resulting in greater details about the witnessed events,
but no association with PTSD symptoms.
As hypothesized, the results also indicated that body state words and word
count mediated the association between experienced violence in 1994 and PTSD
symptoms 14 years after the genocide. These results suggest that experiencing violence
itself is not directly predictive of PTSD symptoms, but instead PTSD symptoms are
predicted by the way survivors think about and describe the violence they experienced.
While this idea has long been the foundation of cognitive interventions for PTSD, it is
notable that even in a sample that has experienced immense amounts of trauma
including being repeatedly attacked and brutalized, experienced violence only indirectly
predicted PTSD symptoms.

41
While experienced violence was fully mediated by linguistic components of the
genocide testimonies, witnessed violence and witnessed harm to family members
directly predicted PTSD symptoms, even after including linguistic components of the
genocide testimonies. Specifically, witnessing violence directly predicted more
hyperarousal symptoms, and witnessing harm to family members directly predicted less
intrusion symptoms. While witnessing violence positively predicting PTSD symptoms
was consistent with study hypotheses, witnessing harm to family members negatively
predicting PTSD symptoms was not. One possible explanation for this result is that for
orphaned survivors of the genocide, knowledge of what happened to family members
may predict less intrusion symptoms, even if that knowledge derives from having
witnessed violence against family. Indeed, research has found that survivors reported
that learning what happened to their family members through the Gacaca trials, while
painful, was also comforting and beneficial (Longman, 2009; Funkeson, Schroder,
Nzabonimpa, & Holmqvist, 2011).
Genocide characteristics and linguistic components as the primary predictor of PTSD
It might be argued that the genocide exposure variables are the sole
contributors to PTSD symptoms, and if linguistic components were not included in the
model just as much variance would still be explained. In order to address this concern, a
post-hoc exploratory model was run in which the linguistic components were removed
from the model, and the genocide exposure and demographic variables were estimated
to predict intrusion, avoidance, and hyperarousal. Results indicated that the model did
42
not fit the data as well as the final model, χ
2
(14)=17.31, p=.24, CFI=.97, TLI=.96,
RMSEA=.05, SRMR=.14, and the only significant path in the model was witnessing
violence predicting hyperarousal (β=.23, p=.003). Moreover, the amount of variance in
hyperarousal explained by witnessing violence was small and not significant (R-
square=.05, p=.14). Therefore, it seems that the types of words used in the genocide
testimonies still accounted for 26% of the variance in intrusion and 7% of the variance in
hyperarousal after accounting for genocide exposure. In other words, the way survivors
described their genocide experiences was more predictive of PTSD symptoms six years
later, than the genocide experiences were.
Sex and age
The hypotheses that sex would predict negative emotion and that age would
predict somatosensory and perceptual words were not supported. Sex was not
associated with any variable in the model, and although age significantly predicted
sadness, the amount of variance in sadness that was accounted for was not significant
(R-square=.04, p=.24). Age was significantly correlated with witnessed harm to family,
such that older children were less likely to witness harm to family members. During the
genocide, many older children may have been away at school, or if they were home
during the genocide, they may have fled in different directions than their parents.
Consequently families were scattered (Geltman & Stover, 1997), and it appears that
older children were less likely to witness their family members being harmed.
Prediction of three components of PTSD
43
This study assessed the prediction of the three components of PTSD separately,
as cognitive models suggest that they will be differentially predicted by linguistic
components of trauma narratives. However, it may be that it is not necessary to divide
PTSD into its three component parts if the model would fit equally well or better when
PTSD is analyzed as one composite or as a latent variable derived from the three
components. To explore this question, two posthoc models were run. The first was a
composite posthoc model using the overall mean IES-R score as the only outcome in the
model, and the second was a latent variable model in which Avoidance, Hyperarousal,
and Intrusion predicted an overall PTSD latent variable, which was the only outcome in
that model.
The composite model did not fit the data well at all, and in fact, did not
converge, indicating that predicting PTSD symptoms as a composite, in which the three
parts contributed equally to PTSD, did not fit the data. The latent model also did not fit
the data because Intrusion and Hyperarousal were too highly intercorrelated and were
therefore linearly dependent on each other and could not be included as predictors of
one latent variable. However, when Intrusion and Avoidance or Hyperarousal and
Avoidance were included as two predictors of a latent PTSD variable, Avoidance was not
a significant predictor of the latent PTSD variable. Therefore, it appears that for these
data, Intrusion and Hyperarousal are very similar constructs, and Avoidance is different
from both of them.
44
The inability of the composite model and the latent model using all three
composites to fit the data supports a model that predicts multiple PTSD outcomes.
However, it may be that having two outcomes (Intrusion/Hyperarousal and Avoidance)
would fit the data well. As this study was particularly interested in linguistic
components predicting Intrusion symptoms, an exploratory posthoc model was run in
which Hyperarousal was dropped from the model and Intrusion and Avoidance were
two separate, but correlated outcomes. The exploratory posthoc model fit the data
well, χ
2
(95)=99.20, p=.36, RMSEA=.02, CFI=.98, TLI=.98, SRMR=.11. The model almost
exactly replicated the final model, as the only path predicting Intrusion or Avoidance
that was significant in the final model but not this exploratory posthoc model was
Hearing predicting Intrusion, and that path almost reached significance (β=-.17, p=.06).
However, when Hyperarousal was not included, the explained variance in Intrusion
dropped and was no longer significant (R
2
=.20, p=.10). This difference appears to be
due to the loss of the path from Hearing predicting Intrusion, because when that path is
included, the explained variance in Intrusion is comparable to that of the final model
and is significant (R
2
=.28, p=.03). Overall the results support the hypothesis that PTSD
should be broken into component parts when it is being predicted by linguistic
components of trauma narratives.
Implications for Research and Interventions
As genocides continue to destroy lives and decimate societies, survivors are
increasingly called upon to provide their testimony as a way of documenting the
45
atrocities they experienced. Many are answering the call. Although survivors may give
testimony for reasons beyond themselves, such as honoring those who were killed and
ensuring that the world remembers the horrors they experienced, the results of the
present study indicate that the way survivors described their genocide experiences
predicted PTSD (particularly intrusion) symptoms, six years after giving their testimony.
Moreover, the linguistic components of genocide testimonies predicted PTSD symptoms
over and above that predicted by the genocide experiences themselves. These results
are important because they offer a possible option for identifying survivors at the
highest risk of developing PTSD symptoms even among a group of survivors who have
arguably suffered some of the most severe genocide experiences. Moreover, analyzing
testimonies given years ago may still provide insight into current PTSD symptoms.
It is important to note that PTSD symptoms were normally distributed in this
sample, and although the symptoms were very high on average, the way that survivors
conceived of, understood, and ultimately described their experiences contributed just as
much, if not more, to PTSD symptoms than actual genocide exposure. This finding
supports cognitive models of PTSD that argue that the PTSD develops and persists
because of disrupted memory, and suggests that efforts to alter the way participants
think about and describe their genocide experiences, such as that done in trauma-
focused cognitive therapies (Bisson & Andrew, 2003; Van Etten & Taylor, 1998), may
improve PTSD symptoms in genocide survivors. In fact, one study has found that
Narrative Exposure Therapy, a cognitive therapy that asks survivors to recount their
46
genocide experiences in a safe and supportive setting, was successful in reducing PTSD
symptoms in Rwandan genocide survivors (Schaal, Elbert, & Neuner, 2009).
Although the variance in PTSD symptoms that was explained by linguistic
components of genocide testimonies was meaningful, even after accounting for
genocide experiences, a great deal of variance remained. Survivors described their
genocide testimonies without prompting for specific content and without the assistance
of counselors or mental health professionals, suggesting that differences in testimonies
were primarily the result of internal processes. Further understanding of these internal
processes may assist with identification of at risk individuals and the development of
interventions. For example, more research is needed to identify other factors that may
influence survivors’ understanding, and subsequent description, of their genocide
experience, such as coping skills, thinking styles, comfort with emotional expression,
and personality differences. Moreover, research is also needed to determine if external
factors such as social support, poverty, daily stressors, and post-genocide traumatic
events such as sexual assault and domestic violence, are also contributing to PTSD
symptoms in genocide survivors (Miller & Rasmussen, 2010).
Finally, analyzing the genocide testimonies and their associated outcomes using
a mixed-methods approach integrating qualitative and quantitative analytic strategies
would allow for a richer and more in depth understanding of the survivors’ genocide
experiences and genocide testimony development, and how those factors contribute to
mental health outcomes. For example, qualitative analyses could be used to identify
47
important themes such as meaning-making, understanding of the experience, the role
of relationships between survivors, victims, and perpetrators, and quantitative analyses
could assist with linking those themes to mental health outcomes.
Limitations and Strengths
Accurate measurement of the study variables was difficult. For example, the
measurement of the LIWC-derived variables was dependent on the accuracy of the
translations of the 2002 genocide testimonies from Kinyarwanda to English. The 2002
genocide testimonies were all translated by one person. It may be that having one
translator reduced the variance in the types of words used, as one individual may have a
bias towards translating words in certain ways. This possible bias may have decreased
the validity of results, while perhaps also decreasing the variance, and therefore
increasing the likelihood of identifying associations. However, overall, the error
introduced into the accuracy of the genocide testimonies would be expected to reduce
the significant effects in the model. Despite this expected error, the words used in the
translated genocide testimonies still predicted PTSD outcomes and were also predicted
by genocide testimonies. If the translations were perfectly accurate, the associations
that were identified in the models would be expected to be stronger. Additionally, since
the genocide exposure variables and words used in the genocide testimonies were both
derived from the genocide testimonies, the significant associations between them may
be inflated. Use of an independent report of genocide experiences may lead to a
different pattern of relations.
48
Moreover, the nature of the genocide testimonies required a coding system that
may have underestimated the true genocide experiences of the participants. Although
coding of the genocide testimonies may have resulted in the reporting of genocide
exposure in a more naturalistic and personal way than a checklist approach, the
genocide experiences were reported retrospectively in an unstructured manner, and are
therefore limited to information participants chose to share with limited prompting, and
are not necessarily representative of their full genocide experiences. In addition, the
items that comprise each of the three genocide exposure subscales: witnessed violence,
experienced violence, and witnessed harm to family may not contribute equally to
outcomes, but were weighted equally on the subscales (i.e., each type of incident
received a score of one). For example, being threatened to be killed may not predict
traumatic stress to the same degree that being raped might. However, the frequency of
endorsement of items resulted in a few items being endorsed by almost all participants,
and the rest being endorsed by a very small numbers of participants. As a result,
variables that may be particularly salient to PTSD could not be analyzed separately (for
example, rape was endorsed by only 6% of the sample).
The present study also had statistical limitations due to the small sample size
which were further exacerbated by moderate retention rates, a sizable amount of
missing data, and a model with many variables. As a result, the study is underpowered
and parameter estimates may not be as accurate they would be in a larger study.
Although auxiliary variables were included to assist with stabilizing parameter estimates
49
and improving power, the study does have more variables than are typically suggested
for such a small sample size. The original hypothesized model included latent variables
predicted from the linguistic components, but for the most part the linguistic
components were not correlated, and so the linguistic component variables had to be
used independently, resulting in many additional estimated paths. Despite these
statistical limitations, many of the hypothesized paths were significant. It must be
noted that although four of the five reported fit indices were within acceptable limits,
SRMR was above the suggested cut-off. As SRMR is susceptible to inflation in small
sample sizes, the final model was considered within acceptable limits, but nonetheless
some caution may be warranted when interpreting the parameter estimates.
Finally, findings may be limited in their generalizability to other OHH in Rwanda,
as the participants themselves were not a random sample of OHH, but rather were all
members of the AOCM organization who were selected for participation by the AOCM
leadership. Therefore, the participants may not be representative of OHH more
broadly. Moreover, the individuals who participated in the study were all willing to
provide their genocide testimonies, and 99% were willing to have their identities
connected to their testimonies, and may therefore be a unique self-selected group,
which may further limit the generalizability of the results.
Despite these limitations, this study had a number of strengths including having
multiple time points which allowed for prediction of PTSD symptoms from reported
genocide exposure and LIWC variables, and the inclusion of multiple types of genocide
50
exposure and all three aspects of PTSD. In addition, by analyzing genocide testimony
rather than trauma narratives that prompt participants to engage in “narrative reliving”,
the study was able to identify the lexicon used when describing genocide experiences in
a more naturalistic manner. In addition, although genocide exposure was measured in a
systematic checklist-like manner, the open-ended approach allowed participants to
describe their experiences as they chose, which may have resulted in participants
sharing experiences that were more salient to them, or that may not have been
captured with a more traditional checklist approach. In addition, this study was the first
to examine the long-term prediction of PTSD symptoms from linguistic components of
descriptions of traumatic experiences, as previously the longest follow-up after
constructing a trauma narrative was 11 months (Dekel & Bonanno, 2011).
Finally, this study was conducted with the full collaboration and participation of
AOCM, a grass-roots community organization of OHH. AOCM’s participation helped
increase the cultural relevance and sensitivity of the study and ensured that the
questions being asked were meeting the specific needs of their membership. The
extensive development work including the use of focus groups of OHH, feedback on,
and approval of, all study materials by the AOCM leadership, and use of AOCM
members as interviewers all contributed to a study grounded in community-centered
and culturally-appropriate values and research practices.

51
Conclusion
As genocide and ethnic conflict continue to spread around the world, survivors
may be increasingly called upon to give testimony. It may be that their efforts to help
others could also be used to help them. Overall this study was able to identify patterns
of word usage in genocide testimony that significantly predicted PTSD symptoms over
and above the variance predicted by genocide experiences in a sample of genocide
survivors who had extremely severe traumatic experiences. It may be possible that with
further understanding, refinement, and replication, analyses of genocide testimonies
could be used to help identify genocide survivors who are at risk of long-term PTSD
symptoms, even years after they provide their testimonies.

52
Chapter 2: Risk Pathways from 1994 Rwandan Tutsi Genocide Exposure to
Distress and Traumatic Stress in Orphaned Heads of Household
Chapter 2 Abstract
As the nature of war shifts from war between nations to local ethnically-based
conflict, millions of children are growing into adulthood having experienced severe war
and ethnic conflict. Orphaned survivors of the 1994 Rwandan Tutsi Genocide were not
only exposed to extraordinarily severe forms of violence, but many of these children
took on the responsibility of caring and providing for other child survivors. The present
study used structural equation modeling (SEM) to investigate the contributions of
modifiable post-genocide risk factors on mental health outcomes of orphaned heads of
household (OHH) in Rwanda 14 years after the genocide. Participants were 100 OHH
who were members of an OHH community organization. Information on genocide
experiences was collected in 2002 and a follow-up of 61 of the 100 OHH was conducted
in 2008/2009 to assess post-genocide risk factors and mental health. Virtually all OHH
reported witnessing and experiencing violence, having low social support and high levels
of poverty. Many had high rates of traumatic stress and distress. In contrast, 90% had
completed primary school and 46% had completed secondary school. Lack of education
substantially predicted both distress and traumatic stress, lack of resources significantly
predicted lower educational attainment, and social support predicted distress. After
accounting for post-genocide risk factors, genocide experiences still directly predicted
distress and traumatic stress. Results suggest that public health and community efforts
53
to increase resources, improve educational outcomes, and strengthen and expand social
support networks may improve mental health outcomes of OHH.

54
Chapter 2 Introduction
As the nature of war shifts from war between nations to local ethnically-based
conflict, children are increasingly bearing the brunt. It is estimated that 90% of conflict-
related deaths between 1990 and 2005 were of civilians, and approximately 80% of
these were women and children (UNICEF, 2005). Armed conflicts between 1986 and
1996 killed approximately 2 million children and left 4-5 million disabled, 12 million
homeless, more than one million orphaned or separated from their parents, and 10
million psychologically traumatized (UNICEF, 1996). As a result of these conflicts,
millions of children grow into adulthood having experienced severe war and ethnic
conflict as children.
One such group is orphaned child and adolescent survivors of the 1994
Rwandan Tutsi Genocide. Orphaned survivors of the genocide were not only exposed to
extraordinarily severe forms of violence, but their parents were killed, sometimes in
their presence. After the genocide, many of these children took on the responsibility of
caring and providing for other child survivors (Schaal & Elbert, 2006). Consistent with
findings from other conflicts, research has documented that child survivors of the 1994
Rwandan Tutsi Genocide are at high risk of mental health concerns (Dyregrov, Gupta,
Gjestad, & Mukanoheli, 2000; Schaal & Elbert), with 44% of orphaned survivors meeting
full DSM-IV criteria for posttraumatic stress disorder (PTSD) 10 years after the genocide
(Schaal & Elbert). However, the range in PTSD symptoms in orphaned survivors is not
wholly explained by genocide exposure. In fact, after controlling for genocide exposure,
55
researchers have found higher rates of PTSD in children living in orphan headed
households than those living in orphanages (Schaal & Elbert), suggesting that other risk
factors, such as lack of resources, education, and social support may be contributing to
mental health outcomes in this population. Indeed, while 44% of orphaned survivors
were found to have PTSD, 56% were not (Schaal & Elbert, 2006). It seems therefore that
genocide exposure alone does not determine mental health. Knowing this, it is critical
to understand the contributions of modifiable risk factors to mental health outcomes,
over and above that of genocide, because these modifiable post-genocide factors might
be used to positively impact mental health outcomes of orphaned child survivors of
genocide.
To address this gap in the literature, the present study used structural equation
modeling (SEM) to investigate the contributions of post-genocide risk factors on mental
health outcomes of orphaned heads of household (OHH) in Rwanda 14 years after the
genocide. Further, the model tests the hypotheses that post-genocide risk factors are
predicted by genocide experiences and partially mediate the association between
genocide exposure and mental health.
1994 Rwandan Tutsi Genocide
Rwanda, a small country in East Africa, experienced decades of increasing ethnic
tension and violence between the Hutu majority (approximately 84% of the population)
and Tutsi minority (15%) (CIA, The World Factbook, March 6, 2012). On April 6, 1994 a
government-backed campaign to exterminate the Tutsi minority spread throughout
56
Rwanda. Over the next 100 days, one-seventh of the population (approximately
800,000 to 1,000,000 Tutsis and 50,000 moderate or sympathetic Hutus) was brutally
butchered. Death was inflicted by decapitation, stabbing, clubbing, drowning,
starvation, and other horrific methods (Human Rights Watch, 1999). The perpetrators
of the violence were not only military and paramilitary groups, but also neighbors,
former friends, and even family members (Staub, Pearlman, Gubin, & Hagengimana,
2005). Most people were killed in their local communities by people who were known
to them (Dyregrov et al., 2000).
Survivors of the genocide were exposed to extreme levels of physical and
psychological violence including rape, torture, mutilation, and witnessing their family
members and loved ones being brutally attacked and murdered. Studies have reported
that 94% of people in Rwanda during the genocide experienced at least one genocide
event including witnessing the murder of family members, having their property and
homes destroyed, and having their lives threatened (Pham, Weinstein & Longman,
2004). Virtually every child survivor witnessed violence and believed they would die
during the genocide (Dyregrov et al. 2000). Many children lost most, if not all, of their
family members, which contributed to one of the highest rates of double orphans
(children who have lost both parents) in sub-Saharan Africa (Monasch & Boerma, 2004)
and the widespread presence of orphaned heads of household.

57
Genocide exposure and risk factors
Bronfenbrenner's social ecological model of child development (1979) proposes
that aspects of the environment including disrupted social networks, poverty, and lack
of access to institutions such as schools, put children at higher risk for poor mental
health outcomes. Support for these risk factors predicting the mental health of children
affected by armed conflict has been found in several studies (see Betancourt and Khan,
2008 for a review). However, war and armed conflict may generate or exacerbate these
risk factors, resulting in secondary adversities (Saltzman, Layne, Syeinberg, Arslanagic, &
Pynoos, 2003). The social ecological model offers a framework for examining risk
factors and post-genocide mental health of orphaned survivors.
Survivors have identified several daily stressors that were exacerbated by the
genocide. Using three ethnographic methods (free listing, key informant interviews,
and pile sorts), Bolton (2001) interviewed people living in rural areas of Rwanda about
the daily concerns they face. Participants identified that their most pressing concerns
were lack of resources (i.e. poverty, lack of food / land / housing / shelter), disruption of
social networks due to suspicion and lack of trust, and the existence of widows and
orphans. In addition, lack of schools, and thus limited access to education, was also a
significant concern. Finally, survivors reported difficulty coping with post-genocide
mental health concerns including traumatic stress and lack of motivation and hope;
however, survivors rated mental health concerns as less distressing than their lack of
resources and disrupted social networks
58
Orphaned heads of household and risk factors
Lack of resources, diminished social networks, and difficulty attaining education
can be particularly challenging for OHH. A study of 68 orphans in post-genocide Rwanda
found that many felt overwhelmed by daily tasks and the hard labor required to meet
their basic needs, and many regularly went without food (Schaal & Elbert, 2006). A
study of 538 youth heads of household found that 44% reported eating only one meal a
day over the last week, and while most had a blanket, spare set of clothes, and a latrine,
only 10.8% had a mattress (Boris et al., 2008).
Orphans in Rwanda have also been found to have a severe lack of social support,
most notably support from an adult (Thurman et al., 2006). This was especially true for
the OHH, who thought that unlike their younger siblings or housemates, they had no
one to turn to for advice or assistance (Ward & Eyber, 2009). It has been found that
OHH also lack trust in their neighbors and others in the community (Boris, Thurman,
Snider, Spencer, & Brown, 2006; Thurman et al.). One study found that 76% of youth
heads of household agreed strongly or very strongly with the statement that the
community rejects orphans (Boris et al., 2008). Although OHH have few adults to turn
to for help, some do report peer relationships, often with other orphaned survivors,
which are sources of positive support (Thurman et al.; Ward & Eyber; Boris et al., 2008).
Finally, researchers have found that orphans in Rwanda had worse educational
outcomes than their non-orphaned peers, independent of the genocide experience
Akresh & de Walque, 2008). Moreover, children who were in Rwanda at the time of the
59
genocide had worse educational outcomes than those who were not (Akresh & de
Walque). Thus, orphaned survivors of the genocide shoulder a double burden of
experiencing the genocide and being orphans when they are attempting to access
education.
Genocide exposure and mental health
Traumatic stress symptoms related to the genocide are highly prevalent in
genocide survivors (Dyregrov et al., 2000; Pham et al., 2004; Schaal & Elbert, 2006). One
study of 3,030 children who were in Rwanda during the genocide found that 79% had
traumatic stress symptom scores that were comparable to diagnosable PTSD one year
after the genocide (Dyregrov et al.), and a second large epidemiological study of 1,547
children in Rwanda found that 54-62% had probable PTSD diagnoses one year after the
genocide, with 75% of those who experienced the most genocide exposure having
probable PTSD diagnoses (Neugebauer et al., 2009).
In addition to traumatic stress, a substantial proportion of survivors of the 1994
Rwandan Tutsi Genocide have been found to have distress symptoms, including
depression and anxiety. One study of 539 youth heads of household in Rwanda found
that 46.9% had depression scores that exceeded the diagnostic cut-off on the Center for
Epidemiological Studies Depression Scale (Boris et al., 2008). Rates of diagnosed major
depression in adults in Rwanda range from 16% of men and women in a commune in
rural Rwanda (Bolton, Neugebauer, & Ndogoni, 2002) to more than 65% of female
genocide survivors (Cohen, et al., 2009).
60
Differential pathways to specific mental health outcomes
While mental health outcomes, specifically traumatic stress and distress
symptoms often co-occur, they may be differentially predicted by genocide exposure
and risk factors. The one study that investigated different predictors of depression and
traumatic stress found that in adult female survivors, genocide exposure was the most
significant predictor of PTSD, while daily stressors predicted depression (Cohen et al.,
2009). Research on war-affected refugee populations has also found that daily
stressors, such as inadequate social support, poverty, and unemployment predict both
depression and PTSD, whereas exposure to war-related events tends to only predict
PTSD symptoms (Heptinstall, Sethna, & Taylor, 2004; Sack, Clarke, & Seeley 1996;
Schweitzer, Melville, Steel, & Lacherez, 2006). Thus, when developing a model to
predict mental health outcomes from genocide exposure and risk factors, it is important
to allow for differential prediction of distress and traumatic stress.
In addition to differential outcomes, diverse genocide experiences may predict
mental health and risk factors in unique ways and may significantly differ in their
potency (Shaw, 2003; Layne et al, 2010; Ng, unpublished dissertation). Although
research has consistently found a general dose-response association between war
exposure and mental health, less is known about how specific types of violence
exposure differentially predict different types of mental health outcomes, such as
traumatic stress and distress (Layne et al).

61
Present study
This study focused on OHH genocide survivors, a particularly vulnerable
population in post-genocide Rwanda, and improved on the existing literature by testing
associations between multiple forms of genocide exposure, risk factors, and mental
health outcomes in one cohesive model and by using two post-genocide time points (8
and 14 years after the genocide). Advanced statistical methods were used to test
overall model fit and specific indirect effects. While genocide exposure has been
associated with mental health problems, studies have typically collected data on
genocide exposure and mental health functioning concurrently. It is possible that
recalling genocide experiences could color participants’ perception of their current
mental health. A strength of the current study is that genocide exposure was reported
six years prior to reports of mental health and risk factors. This study also seeks to
clarify the etiological importance of different types of genocide exposure for specific
outcomes and the pathways that connect them.
Specifically this study hypothesized that post-genocide risk factors (i.e. lack of
social support, education, and resources) would partially mediate the association
between genocide experiences and traumatic stress, and would fully mediate the
association between genocide experiences and distress, such that genocide experiences
would positively predict post-genocide risk factors and traumatic stress, and in turn, the
risk factors would positively predict distress and traumatic stress. Sex and age were also
included as potentially important covariates. Sex was specifically included as a predictor
62
of distress, as women who have been exposed to violence have consistently been found
to have higher levels of mood and anxiety symptoms than men (Pat-Horenczyk et al.;
Shaw, 2003; Pine & Cohen, 2002). Age was included as a predictor of Lack of Education,
because older individuals would have had more years to accrue educational experience
prior to the genocide. The hypothesized model is pictured in Figure 4.
Figure 4. Chapter 2 hypothesized model.

Note. Paths where a positive association was predicted are represented with a plus sign (+) and paths where a
negative association was predicted are represented with a minus sign (-). All paths between genocide exposure
variables and Lack Social Support, Lack of Education, and Lack of Resources are hypothesized to be positive.

63
Chapter 2 Method
Participants
100 OHH who were members of the Rwandan Association des Orphelins Chefs
de Ménages (AOCM) (i.e. the Association of Orphans Chiefs of Household), participated
in this study in 2002 and 61 of them participated in a 2008/2009 follow-up study. Of the
100 original participants, 58% were male and in 2002 they ranged in age from 13 to 35,
with an average age of 22. In 2002 they were caring for 2 children on average. By
2008/2009, 35% of the participants were married, 25% had biological offspring, and on
average they lived with three other people.
Procedures
The data from this study come from two sources. The first is the AOCM
Genocide Oral History Project which was conducted in 2002 by AOCM, in collaboration
with Dr. Donald Miller of the USC School of Religion and Ms. Lorna Miller of All Saints
Episcopal Church in Pasadena, CA. The second is the AOCM-USC Trauma Project which
was conducted in 2008/2009 by AOCM, Dr. Beth Meyerowitz of the USC Department of
Psychology and Dr. Miller (Meyerowitz et al., 2010). The AOCM-USC Trauma Project
followed up on the participants in the 2002 AOCM Genocide Oral History Project.
Details of the sample, study development and design, and data collection procedures
have been previously reported (Ng, unpublished dissertation). In 2002, approximately 8
years after the genocide, the Genocide Oral History Project used semi-structured
interviews to collect and record 100 genocide testimonies of AOCM members across
64
Rwanda. Interviews were conducted by OHH, in participants’ homes, in locations in the
villages near their homes that were selected by the participants, or in the AOCM
headquarters in Kigali. Genocide testimonies and interviews were conducted in
Kinyarwanda and were translated and transcribed into English by a native Kinyarwanda
speaker. Three of the genocide testimonies could not be located, and so there is no
information about genocide experiences for three of the participants.
The AOCM-USC Trauma Project, a 2008/2009 follow-up study, was conducted to
gather information on mental health and risk and resiliency factors 14 years after the
genocide. Sixty-one of the original 100 participants participated in the 2008/2009
follow-up. Interview questions and measures were selected from existing validated
measures and were developed by the USC researchers in collaboration with the AOCM
board and data from focus groups of OHH. All measures were forward- and back-
translated between English and Kinyarwanda by native Kinyarwanda speakers to ensure
accuracy. Six AOCM members conducted the interviews in Kinyarwanda after a week-
long training. Interviews were audiorecorded and translated by native Kinyarwanda
speakers. Study methods were approved by the University of Southern California
Institutional Review Board and the AOCM board.
Interview Coding
The 2002 and 2008/2009 interviews were coded to obtain quantitative
measurements of genocide experiences and lack of resources in 2008/2009. Interview
coding procedures have been described in detail previously (Ng, unpublished
65
dissertation). Each interview was coded by two of four raters. In cases where
discrepancies or missing information remained even after each coder independently
reviewed her own coding, items were reviewed by the author who made the final
decision based on a close reading of the interview. For the 2002 interviews, out of
4,365 codes (97 narratives and 45 variables), the author made a final decision on 175
(4%). For the 2008/2009 interviews, out of 1,159 codes (61 interviews and 19 variables)
the author made a final decision on 76 (7%).
The 2002 interviews were semi-structured, with only limited prompting (i.e.,
“Tell me about your genocide experience?”), and so items were coded as “Yes” or “Not
mentioned”, and there were no missing data recorded. In contrast, the 2008/2009
interviews were structured and so most items were asked by interviewers and answered
by participants. Thus the 2008/2009 items were coded as “Yes” or “No”, and in the
cases where the question was not asked or answered, information was left as “Missing”.
With the exception of Lack of Education and the Number of People Participants Lived
With in 2008/2009, all coded items were dichotomous. Interrater reliability was
acceptable for all items (Cohen’s Kappa >= 0.50, ICC >= 0.60; Stemler & Tsai, 2008).
Cohen’s Kappa for dichotomous items and Intraclass Correlation Coefficients (ICC) for
individual items are reported in Appendices A and G.

66
Study Measures
Sociodemographics
Participants’ Sex, Age, Number of Immediate Family (prior to the genocide), and
the Number of Children Being Cared for in 2002 were recorded by the 2002
interviewers. Participant Marital Status was recorded by the interviewers on the
2008/2009 questionnaires. The Number of People Participants Lived With In 2008/2009
and whether they had Biological Offspring in 2008/2009 were coded from the
2008/2009 interviews.
Genocide Experiences
Given the possible equifinality of different forms of genocide exposure, recent
guidelines for the use of violence exposure checklists argue that measures should be
divided theoretically to create composite-causal indicators that will allow researchers to
test the pattern of associations of different forms of violence exposure (Netland, 2001;
Netland, 2005; Layne et al, 2010). Drawing on proposed “common denominator”
dimensions of violence exposure (La Greca, Silverman, Vernberg & Roberts, 2002; Layne
et al) and knowledge of the specific types of violence that OHH were exposed to during
the genocide, the following theoretically determined dimensions of genocide exposure
were examined: (1) Experienced Violence, (2) Witnessed Violence, (3) Witnessed Harm
to Family, and (4) Family Deaths. In addition, many OHH had most if not all of their
family members killed. The percentage of family members that were killed may have
had differential effects on risk factors and outcomes than the number of family killed,
67
and therefore (5) Percent of Family Killed was also included as a genocide exposure
variable.
Genocide experiences were coded from the 2002 genocide testimonies. See Ng
(unpublished dissertation) for more information on the development of coding sheets
The coded items were divided theoretically into two sum-scale composites: Witnessed
Violence and Experienced Violence. The Experienced Violence subscale was not
completely parallel to the Witnessed Violence subscale, because some items on the
witnessing violence subscale are impossible to have experienced and survived (e.g.,
witnessing someone being killed). Additionally, one experienced violence item (being
raped with an object) was dropped because no participants endorsed it. The Witnessed
Violence subscale was composed of 22 genocide exposure items and the Experienced
Violence subscale was composed of 19 items.
The third subscale, Witnessed Harm to Family, assessed whether or not
participants mentioned witnessing different immediate family members attacked. Due
to the nature of the descriptions of the genocide narratives, it was often unclear
whether participants were describing the same sibling or family member being attacked
or a different family member, and so coding was limited to four items: mentioning
witnessing an attack on your 1) mother, 2) father, 3) sibling, or 4) other relative, and so
the range in this subscale is limited to 0 to 4 and probably underestimates the number
of immediate family members that participants witnessed being attacked.
68
The 2002 interview also included two self-reported questions about the number
of immediate family members that were alive prior to the genocide (including the
participant) and Family Deaths (the number of family members that were killed during
the genocide). Family Deaths and Percent of Family Killed were also included as types of
genocide experiences. Reliability for all scales was good. ICCs for Witnessed Violence,
Experienced Violence, and Witnessed Harm to Family were .93, .92, and .92,
respectively (see Appendix A).
Risk Variables
Lack of Education.
Lack of Education was coded from the 2008/2009 interview. Participants were
asked about the highest level of education they had attained and responses were coded
into seven ordinal categories: 1) Graduate school (attended and/or completed), 2)
Completed university, 3) Some university, 4) Completed secondary, 5) Some secondary,
6) Completed primary, 7) Less than primary. Lack of Education was treated as a
continuous variable in the analyses. ICC for Educational Attainment was .95.
Lack of Social Support.
Lack of Social Support was measured in 2008/2009 using the reverse scored
mean of an adapted version of the Medical Outcomes Study Social Support Scale (MOS
Social Support Scale: Sherbourne & Stewart, 1991) that was given during the 2008/2009
follow-up. The adapted questionnaire had six items and asked participants to rate their
perceived social support over the past two months on a reverse scored scale from 1 (“A
69
lot”) to 5 (“Not at all”). An example question is “Are there people in your life whom you
can trust?” The Lack of Social Support Scale had a Cronbach’s alpha score of 0.83. See
Appendix H for Lack of Social Support items and their descriptive statistics.
Lack of Resources.
Lack of Resources was measured using a scale created from variables coded from
the 2008/2009 interview that assessed poverty and lack of resources. Nineteen
dichotomous (0=No, 1=Yes) items that were selected as indicators of lack of resources
including not having stable employment, lacking food, not having running water in your
home, and having problems with shelter. In order to improve the internal reliability of
the scale, scale reduction through reliability analysis was used. Lack of Resources was
assessed in 2008/2009, with only 61 of the original 100 participants, so Little’s Missing
Completely at Random (MCAR) test using the EM procedure of SPSS 19’s Missing Values
Analysis procedure was conducted to ensure that scale reduction would not have
resulted in a different scale if the complete sample had been available (Little, 1988).
Results indicated that the Lack of Resources items were not MCAR (Chi-Square = 858.64,
DF=792, p=0.05), and that missingness of Lack of Resources items was negatively
associated with Experienced Violence, indicating that participants who provided
information in 2008/2009 about their Lack of Resources reported significantly less
Experienced Violence during the genocide than participants who did not provide Lack of
Resources data.
70
In order to determine whether there would be a difference in the scales after a
data reduction reliability analysis based on Experienced Violence, the reliability analysis
was run using 1) all participants, 2) participants with low experienced violence (based on
median split), and 3) participants with high experienced violence. Items were considered
to contribute significantly to the internal consistency of the scale if the corrected item-
total correlation score was > 0.20. Seven items contributed to the scales for all groups,
while two items (“Things that can’t be done because of lack of means” and “Sometimes
lacking food”) only contributed to the scale for the total sample and the low
Experienced Violence sample, and three items (“Things that have gotten in the way of
education”, “Problems with shelter”, “Problems with transportation”) only contributed
to the scale for the high Experienced Violence sample. In order to include variance
relevant to both groups, all items that contributed to the scale for any of the three
groups were used to create the Lack of Resources scale. The resulting scale was 12
items, and Lack of Resources was the mean of these twelve dichotomous items. The
Cronbach’s alpha for the complete scale is 0.69. See Appendix H for Lack of Resources
items and their descriptive statistics.
Mental Health
Traumatic Stress.
Traumatic Stress was assessed in 2008/2009 using an adapted version of the
Impact of Events Scale-Revised (IES-R: Weiss & Marmar, 1997). The IES-R is a twenty-
two item self-report questionnaire that measures traumatic stress. Although the IES-R
71
was not developed for use in a Rwandan sample, it has been used in post-genocide
Rwandan samples (Dyregrov et al., 2000). The IES-R asked participants to rate the
distress of each item “with respect to the genocide and your current living situation”
over the past two months. Items are scored from 0 (“not at all”) to 4 (“extremely”). The
scale score was calculated as the mean of the items. There was disagreement between
translators over the translation of Item 17, which was intended to ask whether
participants endorsed “I tried to remove the genocide from my memory,” and so this
item was dropped from the scale, resulting in a 21-item scale. The results of a reliability
test indicate that the Cronbach alpha score for the 21-item scale was .91. See Appendix
H for IES-R items and their descriptives.
Distress.
Distress Symptoms were measured in 2008/2009 using a locally-derived
Symptom Checklist that was developed for this study using responses to focus groups of
OHH to measure different forms of psychological distress in Rwandan OHH, including
depression, anxiety, and behavioral problems. The Symptom Checklist is a 36-item scale
that asks participants to indicate how intense the severity of a particular symptom was
over the past two months. Severity ranged from 0 (“not at all”) to 4 (“extremely”). For
four items, the second set of translators could not agree on the accurate translation,
and so these items were dropped from the scale. Two items were not salient (very low
to no endorsement), and so they were also dropped from the analyses.
72
Scale reduction through reliability analysis was used to improve the internal
reliability of the measure. Since Distress was measured in 2008/2009, with only 61 of
the original 100 participants, Little’s Missing Completely at Random (MCAR) test using
the EM procedure of SPSS 19’s Missing Values Analysis procedure was conducted to
ensure that scale reduction would not have resulted in a different scale if the complete
sample had been available (Little, 1988). Results indicated that the items on the
Distress Symptoms scale were MCAR (Little’s MCAR Test Chi-Sq = 507.12, DF=468,
p=.10), therefore researchers could safely conclude that running a reliability analysis on
the available data and dropping items that did not significantly contribute to the
internal consistency of the scale would produce a scale similar to one if there were no
missing Distress items. Reliability analysis revealed that four additional items had
corrected item-total correlation scores ≤ 0.20 and were dropped from the scale. The
result of this data reduction technique is a 26-item scale with an alpha of 0.91. See
Appendix H for descriptives of Distress items.
Preliminary Analyses
Attrition Analyses
To assess whether any 2002 variables predicted attrition, logistic analyses were
run in which all 2002 variables were included as predictors of 2008/2009 completion.
Using an alpha of p<.10, logistic regression comparisons between subjects retained and
those lost to follow-up revealed that Experienced Violence was the only 2002 variable
that significantly predicted follow-up completion, such that those who experienced
73
more violence were more likely to complete the follow-up interview (p=0.07, mean
Experienced Violence for follow-up completers = 4.29 and for follow-up non-completers
= 3.55). No other 2002 variables significantly predicted follow-up completion. See
Appendix I for attrition analysis results.
Normality of study variables
Study variables were screened for outliers, skewness, and kurtosis. Number of
Children Being Cared for in 2002, Family Deaths, and Witnessed Family Harmed did not
meet the assumption of normality and were transformed. Family Deaths was
successfully transformed using the square root transformation, however, the other two
variables remained non-normal despite use of all transformations. Number of Children
Being Cared for in 2002 and Witnessed Family Harmed were used in their
untransformed form, with the limitation that they are non-normal.
Auxiliary Variables
In order to reduce bias and increase power, auxiliary variables (covariates that
help predict missing values but are not in the tested model) were included in all
analyses (Graham, 2009). To ensure that the model was predicting missing variables as
accurately as possible, variables in the data set that were 1) significantly associated with
dependent variables, 2) were not missing all scores on both the variable of interest and
the auxiliary variable, and 3) were part of the broader study but would not otherwise be
included in the models, were considered auxiliary variables (Enders, 2008). Although
auxiliary variables are most useful when auxiliary variables are highly correlated with
74
the variables in the substantive model (Enders), only one variable was highly correlated
with the study variables, and so a cut-off of r≥0.30 was used to identify auxiliary
variables. See Appendix J for correlations between auxiliary variables and study
variables.
Bivariate correlations between the outcomes of interest, study variables, and
variables in the dataset that were not included in this study, were run to identify
auxiliary variables that met the criteria outlined above. This process identified five
variables that met criteria to be auxiliary variables. Four of the auxiliary variables were
previously examined in Ng (unpublished dissertation): Word Count, Body, Sexual,
Touching, and one of the auxiliary variables (Returned to School by 2002) was coded
from the 2002 interviews but was not used in this study.
Analytic Plan
Path analysis was used to examine associations between the study variables and
consider the mediating role of risk variables in the hypothesized model. Models were
run using SEM with Mplus statistical modeling software (Version 6.12; Muthén &
Muthén, 2011). SEM allows for all of the associations of all the variables to be tested
simultaneously, direct and indirect effects to be identified, the fit of the overall model to
be tested, and the use of auxiliary variables to reduce bias and increase power. In order
to correct for some non-normality found in some variables, the analyses used full
information maximum likelihood (FIML) estimation with robust standard errors (MLR
estimator). Even for small sample sizes, FIML has been found to perform reasonably well
75
(Hoyle & Panter, 1995). In addition, FIML handles missing data, assuming data are
missing at random.
In order to assess the fit of the hypothesized model, the structural path model
shown in Figure 4 was tested wherein all hypothesized paths shown in Figure 4 were
estimated freely and all other possible paths not shown in Figure 4 were fixed at 0
(Model A). Goodness of fit was determined using the comparative fit index (CFI),
Tucker-Lewis index (TLI), root-mean-square error of approximation (RMSEA), and
standardized root-mean-square residual (SRMR). Indicators of acceptable model fit for
small sample sizes are considered to be a CFI > .95, TLI> .95, RMSEA < .06, and SRMR <
.09 (Hu & Bentler, 1999).
After testing the hypothesized model, exploratory analyses were conducted to
determine if the fit of the model could be improved with alternative paths. The
exploratory analyses were assessed through a three-step procedure. First, a just-
identified model was tested with all potential paths in the model shown in Figure 4 left
free to vary. Second, model trimming was conducted on the just-identified model
following procedures recommended by Chou and Bentler (2002). Estimated regression
coefficients were inspected, paths with the smallest z-score were systematically fixed to
zero and the model was computed again, until all paths were significantly different from
zero (z>1.96, p<0.05). Once all regression paths were significant, the covariate matrix
among dependent variables was inspected, and non-significant correlations between
76
dependent variables were also fixed to zero. The model resulting from the exploratory
analysis is Model B.
Chapter 2 Results
See Table 4 for descriptives of study variables and Table 5 for correlations of
study variables.

Table 4. Descriptives of Chapter 2 Study Variables
Measure N Mean or % SD Observed Range Possible Range
Pre-Genocide Sociodemographics
Number of immediate family members pre-genocide 100 8.05 2.03 3 - 13
2002 Sociodemographics
Male 100 58.00%
Age 88 22.25 5.06 13 - 35
Number of children being cared for 100 2.23 1.92 0 - 10
2008/2009 Sociodemographics
Married 51 35.29%
Have biological offspring 52 25.00%
Number of people living with 58 3.14 1.67 0 - 9
1994 Genocide Experiences
Witnessed Violence (# of genocide events reported witnessing) 97 7.09 3.05 0 - 14 0 - 22
Experienced Violence (# of genocide events reported experiencing) 97 3.97 1.91 0 - 9 0 - 19
Witnessed Family Harmed 97 1.27 1.27 0 - 4 0 - 4
Family Deaths (# of immediate family killed) 100 5.22 2.02 2 - 10
Percent of Family Killed 100 64.74% 17.61 22.22 - 90.91
2008/2009 Risk Variables
Lack of Education 48 4.50 1.50 1 - 7 1 - 7
1 = Graduate school 2.08%
2 = Completed university 8.33%
3 = Some university 16.67%
4 = Completed secondary school 16.67%
5 = Some secondary school 33.33%
6= Completed primary school 12.50%
7= Less than primary school
10.42%
Lack of Social Support 61 3.96 0.56 2.17 - 5.00 1 - 5
Lack of Resources (mean of 12 dichotomous items) 61 0.62 0.19 .10 - 1.00 0 - 1
2008/2009 Mental Health
Traumatic Stress 61 2.29 0.85 .00 - 3.90 0 - 4
Distress 61 1.37 0.79 .15 - 3.27 0 - 4
Note. Untransformed variables are presented for ease of interpretation.
77

Table 5. Correlations between Chapter 2 Study Variables
Variable 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1. Male
--- .07 -.11 -.08 -.13 -.33* -.20 -.19 -.09 -.09 -.06 .01 -.22 -.15 -.04
2. Age
--- .32** .30** .33* .45** .11 -.08 -.01 -.29** .19 -.02 -.01 -.04 -.36**
3. # family pre-genocide
--- .35*** .20 .28* .24 .09 -.13 -.06 .68** .02 .00 .11 -.06
4. # of children in 2002 --- .20 .15 .52*** -.09 -.23* -.08 -.16 -.51*** .02 -.03 -.21
5. Married 2008/9 --- .60*** .45*** .09 .24 -.07 .04 -.17 -.03 .07 -.22
6. Have children 2008/9
--- .28* .16 .28* -.02 .14 -.08 .15 .19 -.12
7. # Live with 2008/09
--- -.07 -.17 .03 -.21 -.47*** .03 -.13 -.11
8. Witnessed violence
--- .44*** .34*** .14 .10 .27 .33* .30*
9. Experienced violence
--- .23* .11 .27** .27 .20 .25
10. Witnessed harm to family
--- -.02 .04 .08 .14 .11
11. Traumatic deaths
--- .75*** .17 .15 .22
12. % family killed
--- .20 .09 .34**
13. Lack of Education
--- .06 .48***
14. Lack of Social Support
--- .33*
15. Lack of Resources
---
16. Trauma symptoms

17. Distress symptoms

*p<.0. **p<.01. ***p<.001

78

Table 5 continued
Variable 16 17
1. Male
-.15 -.31*
2. Age
.05 -.06
3. # family pre-genocide
.14 .17
4. # of children in 2002 -.17 -.09
5. Married 2008/9 -.11 -.18
6. Have children 2008/9
.20 .18
7. # Live with 2008/09
-.15 -.24
8. Witnessed violence
.37** .29*
9. Experienced violence
.22 .20
10. Witnessed harm to family
-.03 .19
11. Traumatic deaths
.32* .30*
12. % family killed
.31* .25
13. Lack of Education
.41** .44**
14. Lack of Social Support
.20 .35**
15. Lack of Resources
.34** .39**
16. Trauma symptoms
--- .74***
17. Distress symptoms
---
*p<.0. **p<.01. ***p<.001

79
80
Genocide Experiences
On average, prior to the genocide, participants had 8 family members and 5 of
them were killed during the genocide. 63% of participants reported witnessing at least
one family member being attacked. 30% of participants witnessed their mothers being
attacked and 26% witnessed their fathers being attacked. 42% of participants witnessed
their siblings being attacked. 29% witnessed other family members being attacked.
Participants witnessed an average of 7 of 22 different genocide events and personally
experienced an average of 4 of 19 different genocide events. The most frequently
endorsed Witnessed Violence items were witnessing someone being killed (94%),
witnessing someone being attacked (89%), seeing dead bodies (72%), and witnessing
massacres (many people killed at one time; 58%). The most frequently endorsed
Experienced Violence items were having your house damaged or destroyed (95%), being
threatened to be killed (84%), being attacked or assaulted (62%), and being injured
(41%). See Appendix F for genocide exposure items and their descriptives.
Risk Factors
On average participants endorsed 7 out of 12 Lack of Resources items with 80%
stating that they did not have stable employment, 87% did not have running water in
their homes, 62% did not have electricity at home, and 95% sometimes lacked enough
food to feed themselves and their families. Participant Lack of Education was normally
distributed and ranged from less than primary to graduate school, with approximately
10% having not completed primary school, 44% having completed at least secondary
81
school, and 10% having completed university. Lack of Social Support was high, with 80%
of participants rating themselves as having “none” or “very few” people in their lives
who provide social support (4 or 5 on a 5 point scale).
Mental Health
Mental health symptoms were highly prevalent in the sample. The mean
Traumatic Stress score was 2.29 out of 4, with 82% of participants having scores at or
above 1.5, which has been found to be the IES-R cut-off with the best diagnostic
accuracy for assessing PTSD in Vietnam Veterans (Creamer, Bell & Failla, 2003). Distress
scores were also elevated in the sample, with more than 40% of the sample indicating
that they feel “quite a bit” or “extremely” lonely, isolated, sad and depressed, anxious
and slightly upset, scared, and more than 40% indicating that they experience
headaches and nightmares “quite a bit” or “extremely”.
Model Testing
SEM was used to test the hypothesized model, examining the indirect effects of
different forms of 1994 genocide exposure on 2008/2009 distress and traumatic stress
through the mediators of Lack of Social Support, Lack of Resources, and Lack of
Education. The following path model was tested: Traumatic Stress was regressed on
genocide exposure variables and risk variables. Distress symptoms were regressed only
on risk variables. Risk variables were in turn regressed on genocide exposure variables.
Paths between the genocide exposure variables and mental health outcomes were
estimated allowing for the fit of a partially, rather than fully, mediated model to test the
82
hypothesis that genocide exposure would continue to exert direct effects on outcomes,
and would also be transmitted indirectly via risk factors. Two sets of correlations were
estimated 1) Traumatic Stress with Distress, 2) risk variables with each other. Finally,
Distress was regressed on Sex, and Lack of Education was regressed on Age.
The results of fitting the hypothesized structural path model (Model A) are
shown in Figure 5. Results indicate that the hypothesized model did not fit the data
well, χ
2
(13)=34.45, p=.001, CFI=.83, TLI=.41, RMSEA= .13, SRMR=.08.
Figure 5. Results of Chapter 2 hypothesized model (Model A).

Note. All coefficients are standardized. Paths that are not shown are not significant.
*p<.05. **p<.01. ***p<.001.

83
Results of the exploratory analysis fitting the model using the model trimming
procedure (Model B) are shown in Figure 6. Results indicate that the exploratory model
did fit the data well and all fit indices were within acceptable limits, χ
2
(32)=27.93, p=.67,
CFI=1.00, TLI=1.05, RMSEA=.00, SRMR=09. Since the exploratory model fit the data
better and is more parsimonious, it was considered the final model.
Figure 6. Results of Chapter 2 final model (Model B).

Note. All coefficients are standardized.
*p<.05. **p<.01. ***p<.001.

The final model accounted for 20% of the variance in Traumatic Stress and 35%
of the variance in Distress. The model also accounted for 19% of the variance in Lack of
Education, 19% of the variance in Lack of Resources. The variance in Lack of Social
84
Support that was accounted for (11%) was not significant (p=.06). Indirect effects of
covariates and genocide experiences on Distress and Traumatic Stress via risk variables
were tested. See Table 6 for significant indirect pathways in the final model.
Table 6. Significant Indirect Pathways in Model B (Chapter 2)
Indirect pathways B SE β SE
Traumatic Stress
Effect of Male via Lack of Education -.19* .08 -.12* .05
Distress
Effect of Male via Lack of Education -.23* .09 -.15* .05
Effect of Family Deaths via Lack of Education .19 .10 .11* .06
Effect of Witnessed Violence via Lack of Social Support .02* .01 .07* .03

The study hypotheses were largely supported. In the final model, greater
genocide trauma exposure predicted more post-genocide risk factors. In turn, these risk
factors predicted more traumatic stress and distress 14 years after the genocide. In
addition, consistent with the partial mediation hypotheses, two direct effects between
genocide exposure variables and mental health outcomes emerged: witnessing harm to
immediate family members predicted more distress, and witnessing violence predicted
more traumatic stress.
As hypothesized, distress and traumatic stress were independently predicted
while controlling for the other, despite their very high correlation with each other.
Moreover, the pathways leading to their predictions were different. Specifically the
results indicate that social support may have more of an influence on distress than on
85
traumatic stress symptoms, whereas education is associated with both outcomes to a
fairly equivalent, and relatively strong, degree.
As hypothesized, different forms of genocide exposure predicted different
degrees and types of risk variables and mental health outcomes. Witnessed violence
reported in 2002, did predict traumatic stress six years later in 2008/2009. Witnessed
violence did not directly predict distress, but witnessed harm to family members did.
More family deaths predicted less education but not lack of resources, and the opposite
was true for the percent of family killed. Notably, experienced violence did not have
any direct or indirect effects on distress or traumatic stress.
Results indicated that Distress and Traumatic Stress were significantly lower in
males than in females, and that Lack of Education partially mediated the association
between Sex and Distress, and fully mediated the association between Sex and
Traumatic Stress. Specifically, Males had more education that females, and Education
predicted lower Distress and Traumatic Stress. Lack of Education also fully mediated the
association between Family Deaths and Distress and Traumatic Stress, such that the
more immediate family members who were killed during the genocide, the lower the
educational attainment in 2008/2009, Lack of Education, in turn, positively predicted
Distress and Traumatic Stress. Additionally, Lack of Social Support fully mediated the
association between Witnessed Violence and Distress, such that Witnessed Violence
predicted less Social Support, and less Social Support predicted more Distress. Results
86
suggest that exposure to different forms of genocide experiences positively predicted
risk factors, which in turn positively predicted poor mental health outcomes.
Contrary to hypotheses, there was no direct association between Lack of
Resources and Distress or Traumatic Stress. However, Lack of Resources was correlated
with Lack of Education, suggesting that Lack of Education might mediate the
relationship between Lack of Resources and Distress and/or Traumatic Stress, or Lack of
Resources might in turn mediate the relationship between genocide exposure and Lack
of Education. To investigate these alternative hypotheses, a post-hoc model (Model C)
was computed using the model trimming procedure, in which 1) Distress and Traumatic
Stress were regressed on covariates, genocide exposure, and risk variables, 2) Lack of
Social Support and Lack of Education were regressed on Lack of Resources and
covariates and genocide exposure, and 3) Lack of Resources was regressed on covariates
and genocide exposures. Covariates and genocide exposure variables were correlated,
Lack of Social Support and Lack of Education were correlated, and Distress and
Traumatic Stress were correlated.
Results found that the posthoc model (Model C), shown in Figure 7, fit the data
well, χ
2
(32)=29.43, p=.60, CFI=1.00, TLI=1.03, RMSEA=.00, SRMR=09. As expected,
almost all of the same paths that were significant in Model B were significant in Model
C, and Lack of Education was predicted by Lack of Resources. In addition, Lack of
Education was also predicted by Age. The only path that was no longer significant was
Family Deaths predicting Lack of Education. The variance predicted in the post-hoc
87
Figure 7. Results of Chapter 2 posthoc model (Model C).

Note. All coefficients are standardized.
*p<.05. **p<.01. ***p<.001.

model was comparable to that predicted by the final model (20% of the variance in
Traumatic Stress, 35% of the variance in Distress, 33% of the variance in Lack of
Education, 21% of the variance in Lack of Resources, and a non-significant 11% of
variance in Lack of Social Support), with the exception of Model C explaining more of
the variance in Lack of Education (33% vs. 19%).
Indirect effects of covariates and genocide experiences on Distress and
Traumatic Stress via risk and protective variables were tested. Results of significant
indirect pathways are presented in Table 7.

88
Table 7. Significant Indirect Pathways in Model C (Chapter 2)
Indirect pathways B SE β SE
Traumatic stress
Effect of Male via Lack of Education -.19* .08 .11* .05
Effect of Age via Lack of Resources
and Lack of Education -.01* .005 -.06* .03
Effect of Percent of Family Killed via
Lack of Resources and Lack of Education .002* .001 .05* .02
Distress
Effect of Male via Lack of Education -.23** .09 -.15** .05
Effect of Age via Lack of Resources and Lack of Education -.01** .01 .08** .03
Effect of Percent of Family Killed via Lack of Resources
and Lack of Education .003* .001 .06* .03
Effect of Witnessed Violence via Social Support .02* .01 .07* .03

Only one of the indirect effects found in Model B was not also replicated in Model C:
effect of Family Deaths on Distress via Lack of Education. However, three additional
indirect pathways were identified in Model C: 1) Effect of Age on both Traumatic Stress
and 2) Distress via Lack of Resources and Lack of Education, and 3) Effect of the Percent
of Family Killed on Traumatic Stress via Lack of Resources and Lack of Education. Results
suggest that Lack of Resources and Lack of Education are not simply correlated, but that
genocide exposure predicts Lack of Resources, Lack of Resources in turn predicts Lack of
Education, and Lack of Education in turn predicts Distress and Traumatic Stress.
Although the attrition analyses revealed that only one of the 2002 variables was
associated with completion of the follow-up study (Experienced Violence), there may
have been differences in completers versus non-completers that were not measured in
this study. In order to assess the degree to which data from complete cases differed
from data from incomplete cases, the final model was run using only complete cases
89
and model fit was assessed. The model using only complete cases (N=61) fit the data
just as well as the model with all cases (N=100): χ
2
(32)=25.02, p=.81, CFI=1.00, TLI=1.08,
RMSEA=.00, SRMR=.09, and replicated ten of the eleven significant paths in the final
model.
The only path that was not significant in the model with complete cases was
Male Sex predicting Distress (β=-.11, p=.13). As the beta weight was also small (though
significant) in the final model (β=-.14, p=.05), it may be that the model with complete
cases does not have enough power to detect the effect. In addition, the correlation
between Male Sex and Witnessed Violence was no longer significant when only
complete cases were used (r=-.12, p=.32). This drop in association from r=-.20 to r=-.12
is surprising, given that Sex and Witnessed Violence were both collected in 2002, neither
was associated with missingness, and none of the other associations appear to be
different. Overall however, the model with complete cases appears to replicate the
final model quite well, which supports the model’s validity despite the large amount of
missing data.

90
Chapter 2 Discussion
OHH in post-genocide Rwanda reported high levels of genocide experiences, low
social support, high levels of poverty, and high rates of traumatic stress and distress. In
contrast, 90% of the sample completed primary school (6
th
grade), which is the level of
compulsory education in Rwanda (UNESCO, 2012), while 46% completed secondary
school (12
th
grade). As hypothesized, self-reported genocide experiences positively
predicted post-genocide risk factors 14 years after the genocide, and in turn, these risk
factors predicted worse mental health outcomes over and above the negative effects of
the genocide experiences. For orphaned survivors, the impact of the genocide appears
to have lingering detrimental effects on their mental health directly and indirectly, by
negatively predicting their ability to access resources, support, and education.
Lack of education
As hypothesized, lack of education substantially predicted both distress and
traumatic stress, indicating it may be particularly detrimental for OHH in Rwanda.
Moreover, the association between education and mental health outcomes was not
explained by having more social support (such as through greater access to peer support
in school), as education was not correlated with social support. Nor was the association
explained by having more resources, as lack of resources did not directly predict distress
or traumatic stress. The results instead suggest that education may have a beneficial
impact on distress and traumatic stress independent of these other risk factors.
Notably, when lack of resources was allowed to predict lack of education, genocide
91
exposure no longer directly predicted lack of education. This suggests that genocide
exposure itself does not damage survivors ability or desire to pursue education, and that
as long as resources are accounted for, orphaned survivors successfully pursued
education regardless of their genocide exposure.
Education may be beneficial because it may offer hope and the possibility of a
brighter future to OHH. In addition, going to school is one of the primary normative
experiences for children, and for OHH who are suddenly thrust into the adult roles of
caregiver and breadwinner, education may offer an opportunity for success in a
developmentally appropriate everyday role, as well as providing access to teachers,
school administrators, and other adults. Indeed, education certainly appears to be
extremely important to many of the OHH in this sample, who despite not having parents
and having to care for themselves and other orphaned children, on average achieved
much higher educational levels than similarly aged youth in Rwanda. In 2000, the
proportion of Rwandans aged 6 to 35 who had some primary education was 69.16%,
and the proportion with some secondary schooling was 7.04% (Akresh & de Walque,
2008). Comparatively, in this sample in 2008/2009 89.58% of the OHH had some
primary school education, and 77.08% had some secondary school education.
Moreover in 2008/2009, 36% of the OHH in the study were still in school pursuing more
education. The Rwandan government is very committed to education, and by 2008, the
Rwandan government had begun a plan to increase compulsory education from six to
nine years, including 6 years of primary school, and three years of secondary school
92
(Republic of Rwanda, 2008). Indeed, by 2011, the completion rate of children in primary
school was 78.6% (Republic of Rwanda, 2011). Therefore, it may be that the
participants in this study were encouraged to gain more education than participants in
earlier studies. Nonetheless, the level of education achieved by the OHH in this study
was remarkable.
The difference in education between Rwandan youth overall and the OHH in this
sample may stem in part from the fact that the OHH in this sample were part of AOCM,
a grassroots community organization dedicated to assisting OHH. OHH who were part
of AOCM are a self-selected group who may be more able to access education and
resources than other OHH. Moreover, qualitative reviews of interviews revealed that
many of the OHH mentioned having received assistance with school fees and materials
from the government Genocide Survivors Support and Assistance Fund (FARG), which
provides assistance to genocide survivors. Although the need for FARG far outstrips
resources, it seems that the assistance it provided may have been essential to assisting
the OHH pursue their educational goals, and in turn, that educational attainment may
have resulted in much more positive mental health outcomes years later.
Lack of Resources
While lack of resources is clearly a problem for OHH, it did not directly predict
distress or traumatic stress. Instead, its association with mental health occurred
indirectly, through its prediction of lack of education. In this model, education was the
primary predictor, and lack of resources may only be important to mental health if it
93
impedes educational attainment. It may be that OHH who lack many of the resources
included on the lack of resources scale (i.e., difficulty accessing basic needs including
food, water, transportation, employment, and shelter) are struggling to survive, and
only once these survival needs are met, can they begin to tackle the challenges of
acquiring more education. The threshold for having enough resources to continue
education may not be high. In fact 95% of survivors reported that they lacked food, 42%
regularly ate less than two meals per day, only 20% were employed, and only 12% had
access to water in their homes, however 90% completed primary school and 44%
completed secondary school. The results suggest that even modest improvements in
resources may improve educational outcomes, and these may in turn improve mental
health outcomes.
Lack of resources was significantly predicted by the percentage of family killed,
such that the higher the percentage of family members killed, the fewer resources the
OHH reported having. Although the OHH were the often the oldest surviving children in
the family, many of their siblings were close to them in age and may have contributed to
the household resources through helping with farming, completing odd jobs, gathering
water, cooking, and helping to rebuild shelters. OHH who had fewer surviving family
members may have had less assistance with meeting the basic needs of the family,
resulting in less resources overall.

94
Lack of social support
Social support was low for almost all participants, and was not significantly
correlated with lack of education or resources, indicating that regardless of resources or
educational attainment, all of the OHH in the sample reported having little to no
support. Despite this limited range, social support was still a highly significant predictor
of distress. It seems then that OHH who reported “A little” social support had
significantly less distress than those who reported “None”. Lack of social support
measured in this study refers to the perception of having no one to turn to when things
bother you, if you are sick, if you are need advice or help making decisions, and not
having someone around whom you trust, feel close too, and who understands your
feelings. OHH who have even one person in their life who provides this support may
indeed be much better off than those who report having no one and being totally
isolated. Results also indicate that lack of social support fully mediates the association
between witnessed violence and distress. It may be that witnessing people committing
heinous acts against others leads to a deep sense of distrust and insecurity which leads
to a lack of social support, which in turn predicts distress.
Genocide experiences
After accounting for post-genocide risk factors, non-modifiable genocide
experiences also predicted distress and traumatic stress. Specifically, witnessing harm
to family members positively predicted distress while witnessing violence positively
predicted traumatic stress. Although these genocide experiences are similar and
95
statistically related, these results suggest that witnessing harm to family members is
qualitatively different than witnessing violent acts generally. Witnessing violence would
be expected to lead to feelings of horror, helplessness, and fear, all of which are
associated with traumatic stress, while witnessing family members being harmed would
also lead to feelings of anger, sadness, despair, and depression, which are associated
with distress. It may be that by including witnessed violence and witnessed harm to
family member as correlated predictors of distress and traumatic stress, witnessed
violence accounted for the feelings of horror, helplessness and fear, which would be
present in a witnessing violence regardless of who the victim is, while witnessed harm
to family accounted for the feelings of despair, anger, and sadness that are specific to
witnessing a loved one being hurt.
Contrary to the hypotheses, experienced violence did not predict any of the risk
factors or mental health outcomes in the model. Posthoc analyses revealed that when
experienced violence was the only predictor in a model with distress and traumatic
stress as outcomes, it significantly predicted both distress and traumatic stress (β= .28**
and .30**, respectively). However, when witnessed violence was included as a
predictor in the same model, experienced violence no longer significantly predicted
either outcome (β= .19 and .17 respectively). It seems therefore, that experienced
violence is associated with distress and traumatic stress, but other genocide exposure
variables are more salient in this sample. This result may be due to the fact that the
OHH reported witnessing more incidents (mean=7.09) than they experienced
96
(mean=3.97), and since poor mental health outcomes have been found to have a dose-
response relationship with exposure to traumatic events (see Johnson & Thompson,
2008 for a review), it may be that the sheer number of incidents that were witnessed
overwhelms the effect of the number of incidents that were experienced.
Sex and age
As hypothesized, women had more distress than men. In addition, sex appears
to have an indirect effect on both distress and traumatic stress such that men had less
distress and traumatic stress and this may be because they had more education than
women. The finding that men had more education than women was unexpected, as
research has found equal rates of education in boys and girls in Rwanda (Akresh & de
Walque, 2008). However, as noted above, this sample had much higher educational
attainment than that found in epidemiological studies in Rwanda (Akresh & de Walque),
and it may be that sex differences in education occur at the secondary and university
level, rather than the primary school level. The women in this sample were also
significantly more likely to have children than men, and perhaps the additional
caregiving responsibilities contributed to their lower educational attainment,
particularly at the secondary and university levels. Indeed, the Rwandan Ministry of
Education reported that in 2011, although girls and boys were enrolled in school in
roughly equal numbers, men were more likely to be enrolled in school starting in upper
secondary (10
th
through 12
th
grades) and this disparity increased through postsecondary
education, particularly at public universities (Republic of Rwanda, 2011).
97
Age predicted lack of resources, such that younger OHH had fewer resources
than older OHH. Although all of the OHH in this study had become adults by 2008/2009,
during the genocide they ranged in age from 5 to 27 with a mean age of 14.25. OHH
who were young when they were orphaned may have had a much harder time meeting
their basic needs. For example, it would be reasonable to think that an eight year old
would have a harder time rebuilding a home, farming a field, cooking food, gathering
water, and finding employment compared to an eighteen year old. In addition, an eight
year old with younger siblings aged six and four would be expected to have less
assistance from the siblings with accessing resources than an eighteen year old with
sixteen and fourteen year old siblings. Therefore, it may be that age differences in
resources stem in part from the fact that some OHH were orphaned at much younger
ages than others.
Differential prediction of distress and traumatic stress
As hypothesized, distress and traumatic stress were differentially predicted in
the model, however since distress and traumatic stress were highly correlated, it may be
that having one mental health outcome composed of distress and traumatic stress
would fit the data just as well. To test this hypothesis, an exploratory posthoc model
was run in which a composite outcome consisting of the mean score of the standardized
scores on traumatic stress and distress was used as the only outcome. A composite
score was used rather than a latent score because distress and traumatic stress were
linearly dependent on each other and could not form a latent variable. The results
98
indicated that the posthoc exploratory model did not fit the data well, χ
2
(25)=35.99,
p=.07, CFI=.84, TLI=.78, RMSEA=.07, SRMR=.11. Despite the high correlation between
distress and traumatic stress, the final model in which they both were predicted fit the
data much better than the model in which they were collapsed into a single outcome.
Directionality
While this study improves on others by including two time points and having
participants report their genocide experiences years before they reported on their
mental health and post-genocide risk factors, the risk factors and mental health
assessments were still conducted at the same time. Therefore, no directionality can be
inferred between these variables. It may be that distress and traumatic stress predict
lack of education, social support, and resources, rather than vice versa. Although the
data is cross-sectional, to explore this possibility, a post-hoc model in which distress and
traumatic stress were mediators, and lack of education, resources, and social support
were outcomes was run. The model fit the data well χ
2
(30)=25.71, p=.69, CFI=1.00,
TLI=1.05, RMSEA<.001, SRMR=.08. Results indicated that distress predicted lack of
education (β=.38, p<.001), social support (β=.47, p<.001), and resources (β=.33, p<.001),
however traumatic stress did not predict lack of education or resources, and it predicted
lack of social support in the opposite of the expected direction (β=-.28, p=.05). It may
be that distress and post-genocide risk factors have a reciprocal association, while risk
factors predict traumatic stress, rather than vice versa. This area requires further
99
investigation using longitudinal methods to begin to understand the directional
associations between these variables.
Finally, analyzing the interviews and the mental health outcomes using a mixed-
methods approach integrating qualitative and quantitative analytic strategies would
allow for a richer and more in depth understanding of the how post-genocide risk
factors contributed to mental health outcomes. For example, qualitative analyses could
be used to identify the survivors’ hypotheses about why education and social support
are associated with mental health and then these hypotheses could be tested in future
research.
Implications for Research and Interventions
Around the world, children are being exposed to severe ethnic conflict and
genocide and are growing into adulthood in societies that have been devastated by war.
The results are consistent with those of other studies that have noted that interventions
that target economic, social, and education risk factors may improve mental health
functioning for a substantial number of war-affected people (Miller and Rasmussen,
2010; Miller et al., 2008). These risk factors may be used to identify orphaned survivors
of genocide who are at comparatively higher risk of mental health concerns and may be
the focus of interventions to improve their mental health.
Education appears to play a particularly important role in predicting distress and
traumatic stress years after the genocide. Programs that help OHH access education by
providing transportation, uniform fees, school materials, and assistance with applying to
100
and paying for higher education may substantially reduce distress and traumatic stress.
Efforts to improve access to education for women and girls may be most needed, as the
results indicated that males had higher educational attainment than females. The
results further suggest that efforts to relieve poverty and increase resources, including
providing sustainable employment, increasing agriculture production, rebuilding
shelters, and improving access to clean water, may indirectly improve educational
outcomes, thereby decreasing distress and traumatic stress. Importantly, public health
interventions should attempt to improve access to education in conjunction with
reducing poverty and increasing resources. Targeting only one of these may reduce the
overall effectiveness of the intervention.
The results also suggest that efforts to strengthen and expand social support
networks, such as promoting community groups of others with shared experiences,
developing mentoring networks for OHH, integrating orphans into the community, and
continuing efforts to reconnect fragmented families, such as those currently being
conducted by local community groups and non-profits (i.e., Uyisenga N'Manzi), may also
decrease distress. The goals of social support interventions may not need to be lofty,
because the results suggest that even a small change may be quite beneficial for many
OHH. One example of a psychosocial intervention that targets social support is a large
scale community-based sociotherapy intervention designed to improve social bonding in
post-genocide Rwanda. Researchers found that the distress of participants in the
sociotherapy group was significantly lower post-intervention and at follow-up than
101
participants in a control condition, and the effects were largest for women (Scholte et
al., 2011). This intervention program is owned and carried out by members of the local
population, has been sustainable for over four years, and has involved over 7,000
participants (Scholte et al.). Therefore results are encouraging that interventions such
as these may be culturally-relevant and sustainable options for addressing mental
health needs in post-genocide Rwanda.
For many orphaned survivors of genocide, public health interventions that
improve social support, education and resources may significantly reduce distress and
traumatic stress symptoms, but for some orphaned survivors, particularly those who
have witnessed the most violence, symptom focused mental health interventions may
be warranted. It may be that different types of interventions are needed to address
different types of genocide experiences. For survivors who have witnessed their family
members being harmed, interventions for mood disorders including depression may be
beneficial, while survivors who have witnessed violence generally may require a more
trauma-focused approach to reduce PTSD symptoms. Research on the efficacy of mental
health interventions in Rwanda is limited, but one small study did find that Narrative
Exposure Therapy, a trauma-focused intervention, was effective at reducing PTSD
symptoms in orphaned survivors in Rwanda (Schaal, Elbert, & Neuner, 2009). In
addition, research on children in other war affected areas has found that mental health
treatment was effective at reducing PTSD (Woodside, Santa Barbara, & Benner, 1999;

102
Gordon, Staples, Blyta, & Bytyqi, 2004; Gupta & Zimmer, 2008; Onyut et al., 2005) and
depression symptoms (Bolton et al., 2007).
In addition to increased understanding of the efficacy of western interventions
for depression and traumatic stress, research is needed on the effectiveness of existing
local community-based approaches to addressing risk factors and mental health needs.
Not only may these approaches be more culturally-relevant for genocide survivors in
Rwanda, but they may also be more sustainable than imported interventions. Research
is also needed on predictors of functional impairment in OHH. Functional impairment
was not assessed in this study, but the severity of mental health symptoms did not seem
to correlate with the high functioning of many of the participants, which was most
noticeable in the overall high educational achievement. In a society with such large
mental health concerns, individuals show remarkable resilience and perseverance. It
may be that functional impairment may be a more useful mental health outcome for
determining individuals in most need of the limited mental health resources in post-
genocide Rwanda. Finally, as noted above, more longitudinal research is needed to
address the lack of understanding regarding the directionality of mental health and
other associated variables.
Limitations and Strengths
Information on mental health and risk variables was all collected in 2008/2009,
therefore no directionality between these variables can be inferred. The study also has
moderate retention rates, a sizable amount of missing data, and a small sample size,
103
which may mean the study is underpowered. However, despite these limitations, the
final models fit the data well and many of the hypotheses were supported. Additionally,
findings may be limited in their generalizability to other OHH in Rwanda, as the
participants themselves were not a random sample of OHH, but rather were all
members of the AOCM organization who were selected for participation by the AOCM
leadership. Therefore, the participants may not be representative of OHH more
broadly. Moreover, the individuals who participated in the study were all willing to
provide their genocide testimonies, and 99% were willing to have their identities
connected to their testimonies, and may therefore be a unique self-selected group,
which may further limit the generalizability of the results. The study also used self-
report measures for distress and traumatic stress, which are subject to participant
misunderstanding or biased responding.
In addition to generalizability concerns, the translation process may have
decreased the accuracy of the data. Although measures and interview instruments
were translated and back translated by native Kinyarwanda speakers prior to data
collection, retranslations by a second set of translators indicated a lack of consensus
regarding the correct translation of some of the items. While these specific items were
not included in the analyses, the multiple translations and back translations may have
resulted in slight changes to the wording of items, and may have added additional error
into the data, particularly for the items that were created through coding transcripts.
Although the use of multiple translators may have added error to the data, the coding
104
from the interviews only required accurate translation of concepts rather than exact
words. A comparison of translations indicated that although the exact wording of two
translations of the same interviews differed, the content was the same. Therefore, the
error introduced by difficulties with accurate translation may have been limited.
The nature of the genocide testimonies required a coding system that may have
underestimated the true genocide experiences of the participants. Although coding of
the genocide testimonies may have resulted in the reporting of genocide exposure in a
more naturalistic and personal way than a checklist approach, the genocide experience
were reported retrospectively in an unstructured manner, and are limited to
information participants chose to share with limited prompting, and are not necessarily
representative of their full genocide experiences. Therefore, genocide experiences must
be interpreted as mentioned experiences, rather than complete experiences.
Moreover, it was often difficult to differentiate between different genocide
incidents in the genocide testimonies, and therefore coding was limited to the number
of different types of violence participants reported. If a participant reported witnessing
a beating and later on again referred to witnessing a beating, that participant received a
score of one on witnessed violence for beating, whereas a participant who reported
witnessing a beating and later reported witnessing someone being cut would receive a
score of two on witnessed violence (one for the beating and one for the cutting).
Therefore, the nature of the coding also is expected to underestimate true genocide
experiences.
105
In addition, the items that comprise each of the three genocide exposure
subscales: witnessed violence, experienced violence, and witnessed harm to family may
not contribute equally to outcomes, but were weighted equally on the subscales (i.e.,
each type of incident received a score of one). For example, being threatened to be
killed may not predict traumatic stress to the same degree that being raped might.
However, the frequency of endorsement of items resulted in a few items being
endorsed by almost all participants, and the rest being endorsed by very small numbers
of participants. As a result, variables that may be particularly salient to distress and
traumatic stress could not be analyzed separately (for example, rape was endorsed by
only 6% of the sample).
Despite these limitations, this study provided a more nuanced approach to
predicting distress and traumatic stress in genocide survivors than is typically observed
in post-conflict research by including multiple types of genocide exposure, three
different post-genocide risk factors, and two mental health outcomes in one
comprehensive model. The use of multiple time points also allowed for prediction of
risk and outcome variables from reported genocide exposure, and allowed for a long
separation between the reporting of distressing information about the genocide and the
collection of information on mental health. In addition, although participants were not
asked about their genocide exposure or lack of resources in a systematic checklist-like
manner, the open-ended approach allowed participants to describe their experiences as
they chose, which is a more naturalistic approach, and may have resulted in participants
106
sharing experiences that were more salient to them, or that may not have been
captured with a more traditional checklist approach.
Finally, this study was conducted with the full collaboration and participation of
AOCM, a grass-roots community organization of OHH. AOCM’s participation helped
increase the cultural relevance and sensitivity of the study and ensured that the
questions being asked were meeting the specific needs of their membership. The
extensive development work including the use of focus groups of OHH, feedback on and
approval of all study materials by the AOCM leadership, and use of AOCM members as
interviewers all contributed to a study grounded in community-centered and culturally-
appropriate values and research practices.
Conclusion
In summary, the results provide support for the hypothesis that post-genocide
risk factors partially mediate the association between genocide experiences and mental
health 14 years after the genocide. The results highlight the usefulness of breaking
genocide experiences apart into smaller theoretically determined constructs, as these
constructs predicted risk variables and outcomes in unique and clinically meaningful
ways, investigating multiple mental health outcomes and post-genocide risk factors in
one model, and the identifying key pathways through the risk factors to mental health
outcomes. Knowledge of these pathways may improve the efficacy of interventions
both by identifying variables that can be targeted for change, such as social support,
107
education, and resources, and also by identifying individuals who may have a higher
chance of having more risk factors and poor mental health outcomes.
Results suggest that public health and community efforts to increase resources,
improve educational outcomes, and strengthen and expand social support networks
may improve mental health outcomes of OHH. In addition, secondary and tertiary
interventions such as depression and trauma-focused interventions may be beneficial
for a subset of OHH who have been exposed to comparatively high levels of witnessed
harm to family members and witnessed violence. Taken together, these findings
highlight possible directions for post-genocide psychosocial and public health
intervention programs to improve the mental health of OHH in Rwanda.
108
References

Akresh, R. & de Walque, D. (2008). Armed conflict and schooling: Evidence from the
1994 Rwandan genocide. Policy Research Working Paper, 4606. The World
Bank, Development Research Group, Human Development and Public Services
Team, 1-36.

Ali, T., Dunmore, E., Clark, D., & Ehlers, A. (2002). The role of negative beliefs in
posttraumatic stress disorder: A comparison of assault victims and non-victims.
Behavioural and Cognitive Psychotherapy, 30, 249-257.

Alvarez-Conrad, J., Zoellner, L.A., & Foa, E.B. (2001). Linguistic predictors of trauma
psychology and physical health. Applied Cognitive Psychology, 15, S159-S170.

Amir, N., Stafford, J., Freshman, M.S. & Foa, E.B. (1998). Relationship between trauma
narratives and trauma pathology. Journal of Traumatic Stress, 11, 385-392.

Beaudreau, S.A. (2007). Are trauma narratives unique and do they predict psychological
adjustment? Journal of Traumatic Stress, 20, 353-357.

Betancourt & Khan (2008). The mental health of children affected by armed conflict:
Protective processes and pathways to resilience. International Review of
Psychiatry, 20, 317-328.

Bisson, J. & Andrew, M. (2007). Psychological treatment of post-traumatic stress
disorder (PTSD). Cochrane Database of Systematic Reviews, 3.

Bolton, P. (2001). Local perceptions of the mental health effects of the Rwandan
genocide. Journal of Nervous and Mental Disease, 189, 243–248.

Bolton, P. Betancourt, T., Speelman, L., Onyango, G., Clougherty, K.F., Neugebauer, R.,
Murray, L., & Verdeli, H. (2007). Interventions for depression symptoms among
adolescent survivors of war and displacement in northern Uganda: a randomized
controlled trial. JAMA, 298, 519-527.

Bolton, P., Neugebauer, R., & Ndogoni, L. (2002). Prevalence of Depressive Symptoms in
rural Rwanda based on symptom and functional criteria. Journal of Nervous and
Mental Disorders, 190, 631-637.

109
Boris, N.W., Brown, L.A., Thurman, T.R., Rice, J.C., Snider, L.M., Ntaganira, J., &
Nyirazinyoye, L.N. (2008). Depressive symptoms in youth heads of household in
Rwanda. Archives of Pediatric and Adolescent Medicine, 162, 836-843.

Boris, N.W., Thurman, T.R., Snider, L., Spencer, E., & Brown, L. (2006). Infants and young
children living in youth headed households in Rwanda: Implications of emerging
data. Infant Mental Health Journal, 27, 584-602.

Brewin, C.R., Dalgleish, T., & Joseph, S. (1996). A dual representation theory of
posttraumatic stress disorder. Psychological Review, 103, 670-686.

Bronfenbrenner, U. (1979). The ecology of human development. Cambridge, MA:
Harvard University Press.

Central Intelligence Agency World Factbook (March 6, 2012). Rwanda.

Chou, C., & Bentler, P. (2002). Model modification in structural equation modeling by
imposing constraints. Computational Statistics & Data Analysis, 41, 271–287.

Cohen, M.H., Fabri, M., Cai, X., Shi, Q., Hoover, D.R., Binagwaho, A., et al. (2009).
Prevalence and predictors of posttraumatic stress disorder and Depressive
Symptoms in HIV-infected and at-risk Rwandan women. Journal of Women’s
Health, 18, 1783-1791.

Creamer, M., Bell, R., & Failla, S. (2003). Psychometric properties of the Impact of Event
Scale-Revised. Behavior Research and Therapy. 41, 1489-96.

Dekel, S. & Bonanno, G.A. (2011). Changes in trauma memory and patterns of
posttraumatic stress. Psychological Trauma: Theory, Research, Practice, and
Policy. Advance online publication, 1-9.

Dyregrov, A., Gupta, L., Gjestand, R., & Mukanoheli, E. (2000). Trauma exposure and
psychological reactions to genocide among Rwandan children. Journal of
Traumatic Stress, 13, 3-21.

Ehlers, A. & Clark, D.M. (2000). A cognitive model of posttraumatic stress disorder.
Behaviour Research and Therapy, 38, 319-345.

Ehlers, A. & Steil, R. (1995). Maintenance of intrusive memories in posttraumatic stress
disorder: A cognitive approach. Behavioural and Cognitive Psychotherapy, 23,
217-249.
110
Eid, J., Johnsen, B.H., & Saus, E. (2005). Trauma narratives and emotional processing.
Scandinavian Journal of Psychology, 46, 503-510.

Enders, C.K. (2008). A note on the use of missing auxiliary variables in full information
maximum likelihood-based structural equation models. Structural Equation
Modeling, 15, 434-448.

Fivush, R. (1998). Children’s recollections of traumatic and nontraumatic events.
Development and Psychopathology, 10, 699-716.

Foa, E.B., Molnar, C., & Cashman, L. (1995). Change in rape narratives during exposure
therapy for posttraumatic stress disorder. Jrnl of Traumatic Stress, 8, 675-690.

Foa, E.B. & Riggs, D.S. (1993). Post-traumatic stress disorder in rape victims. In J.
Oldham, M.B. Riba & A. Tasman, Annual Review of Psychiatry, Vol. 12 (pp. 273-
303). Washington, DC: American Psychiatric Association.

Frattaroli, J. (2006). Experimental disclosure and its moderators: A meta-analysis.
Psychological Bulletin, 132, 823-865.

Funkeson, U., Schroder, E., Nzabonimpa, J., & Holmqvist, R. (2011). Witnesses to
genocide: Experiences of witnessing in the Rwandan Gacaca courts. Peace and
Conflict: Journal of Peach Psychology, 17, 367-388.

Geltman, P. & Stover, E. (1997). Genocide and the plight of children in Rwanda. JAMA,
277, 289-294.

Gordon, J.S., Staples, J.K., Blyta, A., & Bytyqi, M. (2004). Treatment of post-traumatic
stress disorder in postwar Kosovo high school students using mind-body skills
groups: A pilot study. Journal of Traumatic Stress, 17, 143–147.

Graham, J.W. (2009). Missing data analysis: Making it work in the real world. Annual
Review of Psychology, 60, 549-576.

Gupta, L., & Zimmer, C. (2008). Psychosocial interventions for war affected children in
Sierra Leone. British Journal of Psychiatry, 192, 212–216.

Hellawell, S.J. & Brewin, C.R. (2004). A comparison of flashbacks and ordinary
autobiographical memories of trauma: Content and language. Behaviour
Research and Therapy, 42, 1-12.
111
Heptinstall, E., Sethna, V., & Taylor, E. (2004). PTSD and Depressive Symptoms in
refugee children. Associations with pre-migration trauma and post-migration
stress. European Child and Adolescent Psychiatry, 13, 373-380.

Hoyle, R.H., & Panter, A.T. (1995). Structural equation modeling: Concepts, issues, and
applications. Thousand Oaks, CA: Sage.

Hu, L., & Bentler, P. (1999). Cutoff criteria for fit indexes in covariance structure analysis:
Conventional criteria vs new alternatives. Structural Equation Modeling, 6, 1–55.

Human Rights Watch (1999). Leave None to Tell the Story: Genocide in Rwanda.

Jaycox, L.H., Zoellner, L., & Foa, E.B. (2002). Cognitive-behavior therapy for PTSD in rape
survivors. Journal of Clinical Psychology, 58, 891-906.

Johnson, H. & Thompson, A. (2008). The development and maintenance of post-
traumatic stress disorder (PTSD) in civilian adult survivors of war trauma and
torture: A review. Clinical Psychology Review, 28, 36-47.

Jones, C., Harvey, A.G., & Brewin, C.R. (2007). The organization and content of trauma
memories in survivors of road traffic accidents. Behaviour Research and
Therapy, 45, 151-162.

Kigali Memorial Center (2012).
http://www.kigalimemorialcentre.org/old/genocide/index.html.

La Greca, A. M., Silverman, W. K., Vernberg, E. M., & Roberts, M. C. (2002). Helping
children cope with disasters and terrorism. Washington, DC: APA

Layne, C.M., Olson, J.A., Baker, A., Legerski, J., Isakson, B., Pašalić, A., et al. (2010).
Unpacking trauma exposure risk factors and differential pathways of influence:
Predicting postwar mental distress in Bosnian adolescents. Child Development,
81, 1053-1076.

Little, R.J.A. (1988). A test of missing completely at random for multivariate data with
missing values. Journal of the American Statistical Association, 83, 1198-1202.

Longman, T. (2009). An assessment of Rwanda’s Gacaca courts. Peace Review, 3, 304-312.

112
Marsh, H.W., Hau, K., & Wen, Z. (2004). In search of golden rules: Comment on
hypothesis-testing approaches to setting cutoff values for fit indexes and
dangers of overgeneralizing Hu and Bentler’s (1999) findings. Structural
Equation Modeling, 11, 320-341.

Meyerowitz, B.E., Taylor-Ford, M., Kurita, K., Ng, L.C., Margolin, G., & Miller, D.E. (2010,
August). Trauma and resilience among orphaned survivors of the Rwandan Tutsi
Genocide. Paper presented at the American Psychological Association Annual
Convention, San Diego, CA.

Miller, K.E. & Rasmussen, A. (2010). War exposure, daily stressors, and mental health in
conflict and post-conflict settings: Bridging the divide between trauma-focused
and psychosocial frameworks. Social Science & Medicine, 70, 7-16.

Miller, K.E., Omidian, P., Rasmussen, A., Yaqubi, A., & Daudzai, H. (2008). Daily
stressors, war experiences, and mental health in Afghanistan. Transcultural
Psychiatry, 45, 611-638.

Monasch, R. & Boerma, J.T. (2004). Orphanhood and childcare patterns in sub-Saharan
Africa: an analysis of national surveys from 40 countries. AIDS, 18, S55-65.

Muthén, L.K. & Muthén, B.O. (2011). Mplus version 6.1.

Netland, M. (2001). Assessment of exposure to political violence and other potentially
traumatizing events. Journal of Traumatic Stress, 14, 311–326.

Netland, M. (2005). Event-list construction and treatment of exposure data in research
on political violence. Journal of Traumatic Stress, 18, 507–517.

Neuendorf, K. (2001). The content analysis guidebook. Thousand Oaks, CA: Sage
Publications.

Neugebauer, R., Fisher, P.W., Turner, J.B., Yamabe, S., Sarsfield, J.A., & Stehling-Ariza, T.
(2009). Post-traumatic stress reactions among Rwandan children and
adolescents in the early aftermath of genocide. International Journal of
Epidemiology, 38, 1033-1045.

Ng. L.C. (Unpublished dissertation). Linguistic Components of Genocide Testimonies
Predict Trauma Symptoms Years Later.

113
O’Kearney, R. & Perrott, K. (2006). Trauma narratives in posttraumatic stress disorder: A
review. Journal of Traumatic Stress, 19, 81-93.

Onyut, LP., Neuner, F., Schauer, E., Ertl, V., Odenwald, M., Schauer, M., & Elbert, T.
(2005). Narrative exposure therapy as a treatment for child war survivors with
post-traumatic stress disorder: Two case reports and a pilot study in an refugee
settlement. BMC Psychiatry, 5, 7.

Pat-Horenczyk, R., Qasrawi, R., Lesack, R., Haj-Yahia, M., Peled, O., Shaheen, M., et al.
(2009). Posttraumatic symptoms, functional impairment, and coping among
adolescents on both sides of the Israeli–Palestinian conflict: A cross-cultural
approach. Applied Psychology: An International Review, 58, 688–708.

Pennebaker, J.W., Chung, C.K., Ireland, M., Gonzales, A. & Booth, R.J. (2007). The
Development and Psychometric Properties of LIWC2007. Austin, TX: LIWC.net

Pennebaker, J.W. & Seagal, J.D. (1999). Forming a story: The health benefits of
narrative. Journal of Clinical Psychology, 55, 1243-1254.

Pham, P.N., Weinstein, H.M., & Longman, T. (2004). Trauma and PTSD symptoms in
Rwanda: Implications for attitudes toward justice and reconciliation. Journal of
the American Medical Association, 292, 602-612.

Pine, D.S. & Cohen, J.A. (2002). Trauma in children and adolescents: Risk and treatment
of psychiatric sequelae. Biological Psychiatry, 51, 519-531.

Ptacek, J.T., Smith, R.E., & Zanas, J. (2006). Gender, appraisal, and coping: A longitudinal
analysis. Journal of Personality, 60, 747-770.

Republic of Rwanda Ministry of Education (2008). Nine Years Basic Education
Implementation: Fast Track Strategies.

Republic of Rwanda Ministry of Education (2011). 2011 Education Statistics.

Rummel, R.J. (1994). Power, genocide and mass murder. Journal of Peace Research, 31,
1-10.

Sack, W.H., Clarke, G.N., & Seeley, J. (1996). Multiple forms of stress in Cambodian
adolescent refugees. Child Development, 67, 107-116.

114
Saltzman, W.R., Layne, C.M., Syeinberg, A.M., Arslanagic, B., & Pynoos, R.S. (2003).
Developing a culturally and ecologically sound intervention program for youth
exposed to war and terrorism. Child and Adolescent Psychiatric Clinics North
America, 12, 319–342.
Schaal, S. & Elbert, T. (2006) Ten years after the genocide: trauma confrontation and
posttraumatic stress in Rwandan adolescents. Journal of Traumatic Stress, 19,
95-105.

Schaal, Elbert, & Neuner (2009). Narrative Exposure Therapy versus Interpersonal
Psychotherapy. Psychotherapy and Psychosomatics, 78, 298-306.

Scholte, W.F., Verduin, F., Kamperman, A.M., Rutayisire, T., Zwinderman, A.H., &
Stronks, K. (2011). The effect on mental health of a large scale psychosocial
intervention for survivors of mass violence: A quasi-experimental study in
Rwanda. PLoS ONE, 6, e2 1819.

Schweitzer, R., Melville, F., Steel, Z., & Lacherez, P. (2006). Trauma, post-migration
living difficulties, and social support as predictors of psychological adjustment in
resettled Sudanese refugees. Australian and New Zealand Journal of Psychiatry,
40, 179-187.

Shaw, J.A. (2003). Children exposed to war ⁄ terrorism. Clinical Child and Family
Psychology Review, 6, 237–246.

Sherbourne, C.D. & Stewart, A.L. (1991). The MOS social support survey. Social Science
& Medicine, 32, 705-714.

Siegel, D.J. (1995). Memory, trauma and psychotherapy: A cognitive science view.
Journal of Psychotherapy Practice and Research, 4, 92-122.

Staub, E., Pearlman, M.A., Gubin, A. & Hagengimana, A. (2005). Healing, reconciliation,
forgiving and the prevention of violence after genocide or mass killing: An
intervention and its experimental evaluation in Rwanda. Journal of Social &
Clinical Psychology, 24, 297-334.

Stemler, S.E. & Tsai, J. (2008). Best practices in interrater reliability. In Best Practices in
Quantitative Methods, J.W. Osborne (Ed.), Los Angeles: Sage, 29-49.

Tamres, L.K., Janicki, D., & Helgeson, V.S. (2002). Sex differences in coping behavior: A
meta-analytic review and an examination of relative coping. Personality and
Social Psychology Review, 6, 2-30.
115

Thurman, T.R., Snider, L., Boris, N., Kalisa, E., Mugarira, E.N., Ntaganira, J., & Brown, L.
(2006). Psychosocial support and marginalization of youth-headed households in
Rwanda. AIDS Care, 18, 220-229.
UNESCO (2012). Education in Rwanda. http://www.nationmaster.com/country/rw-
rwanda/edu-education.

UNICEF (1996). The State of the World’s Children 1996: Children in War.

UNICEF (2005). The State of the World’s Children 2005: Children Under Threat.

Uyisenga N’Manzi. http://uyisenga.org/.

van der Kolk, B.A. & Fisler, R. (1995). Dissociation and the fragmentary nature of
traumatic memories: Overview and exploratory study. Journal of Traumatic
Stress, 9, 505-525.

Van Etten, M.L. & Taylor, S. (1998). Comparative efficacy of treatments for post-
traumatic stress disorder: A meta-analysis. Clinical Psychology and
Psychotherapy, 5, 126-144.

Ward, L.M. & Eyber, C. (2009). Resiliency of child-headed households in Rwanda.
Intervention, 7, 17-33.

Weine, S.M., Kulenovic, A.D., Pavkovic, I., & Gibbons, R. (1998). Testimony
psychotherapy in Bosnian refugees: A pilot study. American Journal of
Psychiatry, 155, 1720-1726.

Weiss, D.S., & Marmar, C.R. (1997). The Impact of Event Scale – Revised. In J.P. Wilson
and T.R. Keane (Eds.) Assessing Psychological Trauma and PTSD (pp. 399-411).
New York, NY: Guilford Press.

Williams, L.M. (1994). Recall of childhood trauma: A prospective study of women’s
memories of child sexual abuse. Journal of Consulting and Clinical Psychology,
62, 1167-1176.

Woodside, D., Santa Barbara, J., & Benner, D.G. (1999). Psychological trauma and social
healing in Croatia. Medicine, Conflict and Survival, 15, 355–367.

116
Appendix A: Reliability Coefficients for Coded 2002 Data

Weighted mean inter-rater Kappa reliability
coefficients
Item N Mean Min Max
Sociodemographics
Age 86 0.99 0.94 1.00
Age (Intraclass Correlation Coefficient) 86 1.00 1.00 1.00
Return to school after the genocide 91 0.94 0.68 1.00
Experienced Violence
Threatened to be killed 97 0.61 0.00 1.00
Attacked or assaulted 97 0.79 0.51 1.00
Injured 95 0.92 0.73 1.00
Shot 97 0.77 0.00 1.00
Shot but not hit 96 0.77 0.60 1.00
Hit with shells/grenades/mortar fire 96 0.79 0.00 1.00
Shells fired at you but weren't hit 97 0.73 0.27 1.00
Cut 96 0.93 0.64 1.00
Impaled 97 0.79 0.00 1.00
Beaten 97 0.94 0.61 1.00
Amputated 97 1.00 1.00 1.00
Raped by one person 97 0.93 0.64 1.00
Raped by more than one person 97 1.00 1.00 1.00
Held as a sex slave 97 1.00 1.00 1.00
Thrown into a pit 97 0.71 0.00 1.00
Drowned 96 1.00 1.00 1.00
Forced to attack others 96 0.79 0.00 1.00
House destroyed 97 0.78 0.46 1.00
Other Violence 96 0.93 0.64 1.00
Witnessed Violence
Witness someone killed 96 0.81 0.00 1.00
Witness people being massacred 97 0.84 0.47 1.00
See dead bodies 97 0.80 0.66 1.00
Witness someone being attacked 96 0.84 0.48 1.00
Witness people screaming 97 0.51 0.00 1.00
Witness people being shot 96 0.87 0.47 1.00
Witness people being shelled 97 0.79 0.50 1.00
Witness people being cut 97 0.78 0.58 1.00
Witness people being impaled 97 0.58 -0.05 1.00
Witness people beaten 97 0.91 0.68 1.00
Witness people being amputated 97 0.84 0.62 1.00
Witness people being raped by one person 97 0.85 0.75 1.00
Witness people raped by more than one 97 0.82 0.58 1.00
Witness people raped with an object 97 1.00 1.00 1.00
Witness people held as sex slaves 97 0.88 0.60 1.00
Witness people thrown in pits or latrines 97 0.89 0.57 1.00
Witness people being drowned 97 0.87 0.62 1.00
117
Witness babies or children killed 97 0.87 0.69 1.00
Witness people forced to kill others 97 0.89 0.46 1.00
Witness houses destroyed 97 0.77 0.50 1.00
Witness people committing suicide 97 0.89 0.44 1.00
Witness other violence 97 0.63 0.00 1.00
Witnessing Family Being Attacked
Witnessed violence against mother 97 0.92 0.60 1.00
Witnessed violence against father 97 0.90 0.00 1.00
Witnessed violence against siblings 97 0.95 0.86 1.00
Witnessed violence against other relatives 97 0.90 0.73 1.00

Intraclass Correlation Coefficients for Scales
Experienced Violence 97 0.92 0.75 0.98
Witnessed Violence 97 0.93 0.88 0.98
Witnessing Family Being Attacked 97 0.92 0.70 1.00

118
Appendix B: Descriptive Information for LIWC Words
Linguistic Category Examples
# of Words
in Category
Validity:
Judges
Alpha:
Binary/Raw
Word Count

Anxiety Worried, fearful, nervous 91 0.38 .89/.33
Anger Hate, kill, annoyed 184 0.22 .92/.55
Sadness Crying, grief, sad 101 0.07 .91/.45
See View, saw, seen 72

.90/.43
Hear Listen, hearing 51

.89/.37
Touch (called "Feel" in dictionary) feels, touch 75

.88/.26
Body Cheek, hands, spit 180

.93/.45
Death Bury, coffin, kill 62

86/.40
Note. “Words in category” refers to the number of different dictionary words that make up the variable category;
“Validity judges” reflect the simple correlations between judges’ ratings of the category with the LIWC variable
(from Pennebaker & Francis, 1996). “Alphas” refer to the Cronbach alphas for the internal reliability of the specific
words within each category. The binary alphas are computed on the occurrence/non-occurrence of each
dictionary word whereas the raw or uncorrected alphas are based on the percentage of use of each of the category
words within the texts. All alphas were computed on a sample of 2800 randomly selected text files from our
language corpus" (Pennebaker, et al., 2007)

Appendix C: IES-R Subscales and Descriptives

Valid Percent

N Mean SD
0 = Not at
all
1 = A little
bit
2 = A
little 3 = A lot
4 =
Extremely
Intrusion Subscale

IES1: Anything remind you of the genocide 61 2.93 1.09 6.56 1.64 18.03 39.34 34.43
IES3: Things reminded me of the genocide 61 2.72 1.19 8.20 3.28 27.87 29.51 31.15
IES6: Had thoughts about the genocide without my consent 61 2.82 1.23 8.20 4.92 21.31 27.87 37.70
IES9: Seeing pictures of the genocide come back to my thoughts 61 2.85 1.38 11.48 4.92 18.03 18.03 47.54
IES14: Felt as if I was back at that time 60 1.90 1.60 35.00 5.00 16.67 21.67 21.67
IES16: Always thought I was in the genocide / Have the same feelings
as if I was in the genocide 61 2.10 1.47 24.59 8.20 19.67 27.87 19.67
IES20: Had only dreams about the genocide 60 2.25 1.45 16.67 13.33 28.33 11.67 30.00
Hyperarousal Subscale

IES4: Irritable and angry 61 2.49 1.41 16.39 3.28 27.87 19.67 32.79
IES10: Was jumpy and nervous 60 2.00 1.46 28.33 3.33 25.00 26.67 16.67
IES18: Could not pretend/act as if things are ok 61 2.44 1.38 14.75 9.84 19.67 27.87 27.87
IES19: People who reminded me of the genocide gave me sweat,
trouble breathing, nausea 61 1.85 1.61 37.70 1.64 19.67 19.67 21.31
IES21: Felt watchful and on guard 61 1.84 1.66 37.70 6.56 14.75 16.39 24.59
Avoidance Subscale

IES5: Avoided letting myself get angry or anxious 61 2.74 1.40 13.11 4.92 19.67 19.67 42.62
IES7: Didn't understand how the genocide could have happened 61 2.97 1.21 4.92 9.84 14.75 24.59 45.90
IES8: Stayed away from people who reminded me of the genocide 61 1.74 1.55 36.07 6.56 24.59 13.11 19.67
IES11: Tried not to remember the genocide 61 1.34 1.42 45.90 6.56 24.59 13.11 9.84
IES12: Was aware of the consequences of the genocide, but didn't deal
with them 61 2.70 1.36 14.75 1.64 16.39 32.79 34.43
IES13: Feelings about the genocide were kind of numb 60 2.12 1.53 25.00 13.33 8.33 31.67 21.67
IES22: Tried not to think about what I saw 61 1.59 1.49 37.70 11.48 18.03 19.67 13.11
Note. Items 2 and 15 were not included because they were equivalent when translated into Kinyarwanda. Item 17 was dropped because of translation difficulties.
Instead of “I tried to remove the genocide from my memory,” it meant, “I did not try to remove the genocide from my memory.”

119

Appendix D: Chapter 1 Attrition Analyses

2002 Variable N Mean SD p
Male
Completed 61 0.56 0.50
0.57
Not Completed 39 0.62 0.49
Age

Completed 53 21.81 5.31
0.31
Not Completed 35 22.91 4.64
Experienced Violence

Completed 58 4.26 1.87
0.07
Not Completed 39 3.54 1.92
Witnessed Violence

Completed 58 7.19 2.99
0.70
Not Completed 39 6.95 3.16
Witnessed Harm to Family

Completed 58 1.32 1.36
0.57
Not Completed 39 1.18 1.14

120

Appendix E: Chapter 1 Auxiliary Variables and Correlations with Outcomes

Family Deaths % of Family Killed Distress Lack of Resources Lack of Education Health Sexual
Word Count -.04 -.04 .38** .30* .17 -.07 -.10
Body States -.23* -.26* -.24 -.28* -.24 -.22* .06
Death Words .12 .16 .01 -.04 -.10 -.08 .21*
Seeing .00 -.10 -.04 -.08 -.14 .07 -.08
Hearing -.06 -.11 .15 -.02 -.06 .04 .08
Touching -.06 -.06 .17 .12 .22 .12 -.13
Anxiety .01 .04 .05 .22 .02 -.01 -.01
Anger .11 .14 .00 .00 -.27 -.12 .37***
Sadness .01 -.02 .24 -.05 -.04 -.16 -.22*
Intrusion .31* .32* .59*** .36** .25 .11 -.03
Avoidance .14 .12 .58*** .25 .42** .33* -.14
Hyperarousal .32* .25* .74*** .34** .38** .12 .00
121
122
Appendix F: Descriptives of Experienced Violence, Witnessed Violence, and Witnessing
Family Violence Subscales

N=97 % Endorsed "Yes" Kappa
Experienced Violence Items
House destroyed or damaged 94.85 0.78
Threatened to be killed 83.51 0.61
Attacked or assaulted 61.86 0.79
Injured 41.24 0.92
Shot at but not hit 27.84 0.77
Beaten (ex: with sticks, impiri, or ubuhiri) 24.74 0.94
Shells/grenades/mortars fired at you but weren't hit 18.56 0.73
Shot 8.25 0.77
Cut (ex: with machete, hoe, or knife) 9.28 0.93
Raped by one person 6.19 0.93
Experienced other violence* 3.09 0.93
Had body parts amputated 3.09 1.00
Raped by more than one person at once (gang raped) 3.09 1.00
Thrown into a pit or latrine 3.09 0.71
Hit with shells/grenades/mortar fire 2.06 0.79
Drowned (attempted) 2.06 1.00
Held as a sex slave 2.06 1.00
Forced to kill or attack others 1.03 0.79
Impaled (ex: with a stick or spear) 1.03 0.79
Witnessed Violence Items
Witness someone being killed 93.81 0.81
Witness someone being attacked 88.66 0.84
See dead bodies 72.16 0.80
Witness many people killed at once (massacred) 57.73 0.84
Witness people being cut (ex: with machete, hoe, or knife) 57.73 0.78
Witness people beaten (ex: with sticks, impiri, or ubuhiri) 38.14 0.91
Witness people being shot 37.11 0.87
Witness people thrown in pits or latrines 36.08 0.89
Witness houses destroyed 36.08 0.77
Witness people being raped by one person 29.90 0.85
Witness babies or children being killed 24.74 0.87
Witness people being shelled (ex: by grenade or mortar fire) 21.65 0.79
Witness people raped by more than one person at one time (gang
raped)
21.65
0.82
Witness people being impaled (ex: with a stick or spear) 16.49 0.58
Witness people having body parts amputated 16.49 0.84
Witness other violence 14.43 0.63
Witness people held as sex slaves 13.40 0.88
Witness people being drowned 9.28 0.87
Witness people screaming for help 7.22 0.51
Witness people committing suicide 7.22 0.89
Witness people forced to kill others 5.15 0.89
123
Witness people raped with objects 4.12 1.00
Witnessing Family Being Attacked Items
Witnessed violence against siblings 42.27 0.95
Witnessed violence against mother 29.90 0.92
Witnessed violence against other relatives 28.87 0.90
Witnessed violence against father 25.77 0.90
Note. Examples of "Other Violence" includes stoning and being trapped inside a house that was set on
fire

124
Appendix G: Reliability Coefficients for Coded 2008/2009 Data

Weighted mean inter-rater Kappa
reliability coefficients
N Mean Min Max
Number of people you live with 56 0.89 0.55 1.00
Do you have children of your own 50 0.96 0.70 1.00
Are you currently in school 57 1.00 1.00 1.00
How much education have you completed 46 0.79 0.68 1.00
How much education have you completed (ICC) 46 0.95 0.87 1.00
Lack of Resources
Do you state that food is a problem for you 61 0.87 0.62 1.00
Do you state that shelter is a problem for you 61 0.76 0.00 1.00
Do you have water in your home 60 0.87 0.62 1.00
Do you have electricity in your home 55 1.00 1.00 1.00
Do you ever lack food 58 1.00 1.00 1.00
Are you a cultivator/farmer 61 0.93 0.80 1.00
Do you have a permanent job 61 0.85 0.38 1.00
Are you unemployed/not paid 61 0.58 0.04 1.00
Are there things that have gotten in the way
of completing your education 60 0.63 0.19 1.00
Are there things that you would like to do
but can't because of lack of means 61 0.73 -0.25 1.00
Do you state that transportation is a problem for
you 61 0.85 0.56 1.00
Do you eat two or more meals per day 48 0.70 0.00 1.00

Intraclass Correlation Coefficients for
Scales
Lack of Resources 61 0.90 0.78 0.96

Appendix H: Scale Descriptives
Social Support Scale Descriptives
Valid Percent
Item N Mean SD 1 = Not at all 2 = A few 3 = Moderate 4 =Many 5 =Definitely
SS1: You can talk about things that bother you 60 2.07 0.69 15.00 68.33 11.67 5.00 0.00
SS2: Get to know your feelings 61 2.02 0.81 24.59 55.74 13.11 6.56 0.00
SS3: Take care of you if you get sick 61 2.03 0.84 24.59 54.10 16.39 3.28 1.64
SS4: Help you make decisions 60 1.92 0.87 33.33 48.33 13.33 3.33 1.67
SS5: You can trust 61 2.02 0.62 16.39 67.21 14.75 1.64 0.00
SS6: Feel close to always 61 2.16 0.80 16.39 57.38 21.31 3.28 1.64

Lack of Resources Scale Descriptives

Item N % Yes
Things you can't do because you lack the means 60 96.70
Do you sometimes lack food 58 94.80
Mention problems getting food 61 88.50
No water in your home 61 86.90
No permanent job 61 80.30
Things get in the way of completing education 59 71.20
No electricity in your home 55 61.80
Not having any source of income 61 45.90
Typically eating less than two meals a day 52 41.70
Being a cultivator/farmer 61 29.50
Mention problems with shelter 61 26.20
Mention problems with transportation 61 23.00
** Items dropped from Lack of Resources Scale: Not having a temporary job, Mentioning other things that are a problem, Not having security,
Not having medical insurance, mentioning that paying for daily expenses is a problem, mentioning paying for education is a problem, mention
that employment is a problem
125

IES-R Scale Descriptives Valid Percent
Item N Mean SD 0 = Not at all 1 = A little bit 2 = A little 3= A lot 4=Extremely
IES1: Anything remind you of the genocide 61 2.93 1.09 6.56 1.64 18.03 39.34 34.43
IES2: Difficulty Sleeping 61 2.36 1.32 16.39 4.92 24.59 34.43 19.67
IES3: Things reminded me of the genocide 61 2.72 1.19 8.20 3.28 27.87 29.51 31.15
IES4: Irritable and angry 61 2.49 1.41 16.39 3.28 27.87 19.67 32.79
IES5: Avoided letting myself get angry or anxious 61 2.74 1.40 13.11 4.92 19.67 19.67 42.62
IES6: Had thoughts about the genocide without my
consent 61 2.82 1.23 8.20 4.92 21.31 27.87 37.70
IES7: Didn't understand how the genocide could
have happened 61 2.97 1.21 4.92 9.84 14.75 24.59 45.90
IES8: Stayed away from people who reminded me
of the genocide 61 1.74 1.55 36.07 6.56 24.59 13.11 19.67
IES9: Seeing pictures of the genocide come back to
my thoughts 61 2.85 1.38 11.48 4.92 18.03 18.03 47.54
IES10: Was jumpy and nervous 60 2.00 1.46 28.33 3.33 25.00 26.67 16.67
IES11: Tried not to remember the genocide 61 1.34 1.42 45.90 6.56 24.59 13.11 9.84
IES12: Was aware of the consequences of the
genocide, but didn't deal with them 61 2.70 1.36 14.75 1.64 16.39 32.79 34.43
IES13: Feelings about the genocide were kind of
numb 60 2.12 1.53 25.00 13.33 8.33 31.67 21.67
IES14: Felt as if I was back at that time 60 1.90 1.60 35.00 5.00 16.67 21.67 21.67
IES15: Difficulty falling asleep 60 2.43 1.43 18.33 3.33 25.00 23.33 30.00
IES16: Always thought I was in the genocide / Have
the same feelings as if I was in the genocide 61 2.10 1.47 24.59 8.20 19.67 27.87 19.67
IES18: Could not pretend/act as if things are ok 61 2.44 1.38 14.75 9.84 19.67 27.87 27.87
IES19: People who reminded me of the genocide
gave me sweat, trouble breathing, nausea 61 1.85 1.61 37.70 1.64 19.67 19.67 21.31
IES20: Had only dreams about the genocide 60 2.25 1.45 16.67 13.33 28.33 11.67 30.00
IES21: Felt watchful and on guard 61 1.84 1.66 37.70 6.56 14.75 16.39 24.59
IES22: Tried not to think about what I saw 61 1.59 1.49 37.70 11.48 18.03 19.67 13.11
Note. Item 17 was dropped because of translation difficulties. Instead of “I tried to remove the genocide from my memory,” it meant, “I did not try
to remove the genocide from my memory.”
126

Distress Scale Item Descriptives

Valid Percent
Item N Mean SD 0 = Not at all 1 = Slightly 2 = Moderately 3=Quite a bit 4=Extremely
Sx33:Feeling anxious/slightly
upset 61 2.77 1.44 14.75 4.92 13.11 22.95 44.26
Sx31: Sad / Depressed 61 2.67 1.36 9.84 11.48 19.67 19.67 39.34
Sx1: Headaches 61 2.10 1.59 26.23 13.11 13.11 19.67 27.87
Sx3: Nightmares 61 1.95 1.47 26.23 11.48 21.31 22.95 18.03
Sx17: Feeling isolated 61 1.95 1.70 36.07 4.92 18.03 9.84 31.15
Sx34: Feeling scared 61 1.95 1.43 26.23 11.48 16.39 32.79 13.11
Sx16: Feeling lonely 61 1.82 1.62 34.43 14.75 6.56 22.95 21.31
Sx13: Feeling nervous 61 1.70 1.48 32.79 13.11 19.67 19.67 14.75
Sx32: Feeling like crying 60 1.67 1.60 40.00 8.33 16.67 15.00 20.00
Sx2: Stomach pain 61 1.64 1.64 44.26 3.28 18.03 13.11 21.31
Sx4: Difficulty Sleeping 61 1.64 1.52 36.07 13.11 18.03 16.39 16.39
Sx7: Too much anger 60 1.57 1.44 36.67 11.67 21.67 18.33 11.67
Sx10: Forgetful 61 1.36 1.34 37.70 18.03 24.59 9.84 9.84
Sx27: Feeling restless 61 1.36 1.63 52.46 8.20 8.20 13.11 18.03
Sx20: Being unable to talk about
what happened 61 1.21 1.58 55.74 9.84 8.20 9.84 16.39
Sx23: Feeling regretful / guilty 61 1.13 1.38 50.82 13.11 18.03 8.20 9.84
Sx12: Feeling hopeless 61 1.10 1.40 50.82 19.67 8.20 11.48 9.84
Sx22: Feeling life is not worth
living 61 1.03 1.41 57.38 11.48 11.48 9.84 9.84
Sx15: Feeling useless 61 0.92 1.39 60.66 14.75 8.20 4.92 11.48
Sx29: Angry because of little
things 61 0.69 1.15 68.85 6.56 14.75 6.56 3.28
Sx30: Thinking of revenge 61 0.66 1.24 72.13 8.20 9.84 1.64 8.20
Sx35: Feeling ashamed 61 0.61 1.14 72.13 9.84 8.20 4.92 4.92
Sx11: Detached/Push people away 60 0.60 1.18 73.33 10.00 6.67 3.33 6.67
Sx8: Difficulty concentrating 60 0.55 0.98 70.00 13.33 10.00 5.00 1.67
Sx36: Losing your mind 61 0.48 1.07 81.97 0.00 9.84 4.92 3.28
127

Sx18: Feeling unable to help
others 61 0.43 0.90 78.69 6.56 8.20 6.56 0.00
Note. Items 5 (Difficulty staying asleep), 9 (Lacking interest in activities), 19 (Feeling unable to accomplish household tasks), and 21 (Feeling
desperate) were not used due to translation problems.
Items 6 (wetting bed) and 26 (fear of getting HIV tested) not used due to them not fitting in a scale and not being salient enough for the
population.
Items 14 (Not showing emotions, 24 (Taking drugs), 25 (Drinking alcohol), and 28 (Getting into physical fights) were dropped from the
Negative Affect scale because their item-total correlations were <0.20.

128
129
Appendix I: Chapter 2 Attrition Analyses

2002 Variable N Mean SD p
Male
Completed 61 0.56 0.50
0.57
Not Completed 39 0.62 0.49
Age
Completed 53 21.81 5.31
0.31
Not Completed 35 22.91 4.64
# Children Being Cared For
Completed 61 2.38 2.20
0.52
Not Completed 39 2.08 1.36
# Family Pre-Genocide
Completed 61 8.11 2.09
0.69
Not Completed 39 7.95 1.96
# of Family Killed (Sqrt)
Completed 61 2.25 0.45
0.73
Not Completed 39 2.22 0.44
% Family Killed
Completed 61 65.15 17.81
0.77
Not Completed 39 64.10 17.50
Experienced Violence
Completed 58 4.26 1.87
0.07
Not Completed 39 3.54 1.92
Witnessed Violence
Completed 58 7.19 2.99
0.70
Not Completed 39 6.95 3.16
Witnessed Harm to Family
Completed 58 1.32 1.36
0.57
Not Completed 39 1.18 1.14

Appendix J: Chapter 2 Auxiliary Variables and Correlations with Outcomes

Trauma
Symptoms
Distress
Symptoms
Social
Support
Daily
Hassles
Education
Level
LIWC word count .31* .38** -.15 .30* -.17
LIWC body words -.38** -.24 -.03 -.28* .24
LIWC feeling words .30* .17 .03 .12 -.22
LIWC sexual words -.08 -.12 -.04 -.32* .14
Returned to school after the genocide - 2002 -.17 -.28* -.07 .01 .55***
*p<.0. **p<.01. ***p<.001

Note. Correlations in bold meet the .30 cut-off to be considered an auxiliary
variable.

130

Abstract (if available)

Abstract Millions of children grow into adulthood having experienced severe war and ethnic conflict as children. One such group is orphaned child and adolescent survivors of the 1994 Rwandan Tutsi Genocide, in which one-seventh of the Rwandan population was murdered over the course of 100 days. After the genocide, many of these children took on the responsibility of caring and providing for other child survivors. Research has documented that child survivors of the genocide are at increased risk of mental health concerns (Dyregrov, Gupta, Gjestad, & Mukanoheli, 2000

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The link between the expression of negative emotion and future well-being: the role of disclosure modality in an East Asian American population

PDF

Risky behaviors, interpersonal conflict, and their relation to fluctuations in adolescents’ diurnal HPA rhythms

PDF

Family conflict, negative mood, and adolescents' daily problems in school

PDF

Partners of breast cancer survivors: Research participation and understanding of survivors' concerns

PDF

The impact of post-traumatic stress and post-traumatic growth on young adult cancer survivors

PDF

Spouse aggression, depression, and physical health: a multivariate longitudinal study of midlife couples

PDF

Perspective taking behavior and social outcomes in late adolescence

PDF

Post-traumatic growth among high-risk youth: predictors, impact of stressful life events, and relationship with changes in substance use behaviors

PDF

Evaluating the effectiveness of interview-informed synthesized contingency analysis for survivors of traumatic brain injury

PDF

Popularity as a predictor of friendship affiliation in adolescence

PDF

Links between discrimination and sleep difficulties: romantic relationships as risk and resilience factors

PDF

Psychosexual adjustment among low-income Latinas with cervical cancer

PDF

A "second chance" program for vulnerable young adults: a program evaluation and a randomized controlled trial of adjunctive motivational interviewing to improve retention

PDF

Adolescent life stress and the cortisol awakening response: the moderating roles of emotion regulation, attachment, and gender

PDF

Couples’ neuroendocrine activity in response to family conflict discussions: the role of self-reported anger and previous marital aggression

PDF

The limits of unidimensional computerized adaptive tests for polytomous item measures

PDF

The link between maternal depression and adolescent daughters' risk behavior: the mediating and moderating role of family

PDF

Family aggression, prosocial friends, and the risk of dating and friend victimization in late adolescence

PDF

Identifying diverse pathways to cognitive decline in later life using genetic and environmental factors

PDF

Arm lymphedema in a multi-ethnic cohort of female breast cancer survivors

Asset Metadata

Creator Ng, Lauren Christina (author)

Core Title Direct and indirect predictors of traumatic stress and distress in orphaned survivors of the 1994 Rwandan Tutsi genocide

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Psychology

Publication Date 07/17/2012

Defense Date 06/05/2012

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Distress,Genocide,OAI-PMH Harvest,orphan,posttraumatic stress disorder,post-traumatic stress disorder,PTSD,resilience,resiliency,risk,Rwanda

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Meyerowitz, Beth E. (committee chair), Huey, Stanley J., Jr.. (committee member), Margolin, Gayla (committee member), McArdle, John J. (committee member), Miller, Donald Earl (committee member)

Creator Email lauren.c.ng@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-59386

Unique identifier UC11290440

Identifier usctheses-c3-59386 (legacy record id)

Legacy Identifier etd-NgLaurenCh-963.pdf

Dmrecord 59386

Document Type Dissertation

Rights Ng, Lauren Christina

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA