Methods
In March 2010, we undertook a population-based study conducted in Djohong District, Cameroon to better ascertain women’s health needs within the refugee and Cameroonian host communities [
17]. Embedded within the survey instrument were indicators of human security derived from the Leaning-Arie model, assessing three domains of psychosocial stability that suggest individuals and communities are most stable when their core attachments to home, community and the future are intact. The Leaning-Arie model was selected as it was designed to be used for community-level, ongoing measurement of human security that could be used to assess a community’s tendency toward instability and conflict over time [
10]. The domain of sense of attachment to home includes a ‘sustainable sense of home and safety providing identity, recognition, and freedom from fear’ measured as 1) number of years in current location; 2) proportion of the population who own land; and 3) self-reported sense of safety. Sense of attachment to community, the second domain in the model, elaborates the individual’s relationship to their community as ‘a network of constructive social or familial support, providing identity, recognition, participation and autonomy’ [
10], such that sudden changes in economic or political circumstances, or perceived inequalities relative to others that appear not to be improving can potentially precipitate violent behavior. Indicators used to measure attachment to community include 1) measures of dynamic inequality between members of different groups, including income and assets; and 2) subjective sense of ties to and inclusion in one’s community or village. The third domain is a positive grasp of the future which includes ‘the confidence that one’s own capacities, and the external social and political structures that confer meaning and stability, will persist for some indefinite period into the future’ [
10]. This can be measured by an individual’s plans to remain in the current village, raise children in the current village, and plans to plant for future harvests. The text below outlines these three domains and provides examples of how each was specifically studied.
Human security framework: Leaning Arie model
Attachment to Home:
-
Number of years living in this village
-
Does any extended family live in this village?
-
Do you own land?
-
Do you feel safe in this village?
Attachment to Community:
-
Is your home made from mud brick or grass/reeds/sticks/tarp?
-
How much land do you own? How much livestock do you own?
-
Do you feel attached to the community in this village?
-
If you needed medical care, would someone take you to the hospital?
Positive outlook for the Future:
-
Do you want to grow old and die in this village?
-
Will your children be working in this village in 10 years?
-
Will you be using your land in 2 years?
-
Do you plan to stay in this village?
Sampling strategy
Female refugee and host population participants were selected using a two-stage household cluster design. The first stage included a random selection of 40 clusters (villages) weighted by refugee and host population estimates from the UN High Commissioner for Refugees (UNHCR) registration lists of January 2010 and from Djohong District government officials, respectively. The latter figures were 2010 estimates of the Cameroonian population using 2006 census figures adjusted for annual population growth. Since the human security dataset came from a study designed to also characterize gender-based violence (GBV) in the region, the estimated prevalence of GBV was used to determine sample size. Based on an estimated prevalence of 20-30% from previous GBV studies in Africa [
18], a desired precision of 0.05, and a design effect of 2.0, we estimated a sample size between 500 and 650, assuming a non-response rate of 5%. Since observations within a cluster may be more alike than observations across clusters, particularly with shared perceptions of human security, we took this intra-cluster correlation into account in the sample size calculation. This design effect, defined as the ratio of the variance taking into account the cluster sample design and variance of a simple random sample design with the same number of observations, was conservatively estimated at 2.0 based both on previous sexual violence studies [
19,
20]. To minimize the possibility of increased intra-cluster correlation and homogeneity on the outcome of sexual violence prevalence and to limit design effect, we sampled more clusters in the first stage and fewer households in the second stage [
21]; the second stage included 15 randomly selected women per cluster for a final sample size of 600 respondents.
In stage one, clusters were defined as villages. To account for the modest variation in village populations across the district, probability proportional to size (PPS) sampling was used to randomly select clusters. As the refugee population does not live in segregated areas but is interspersed among the host Cameroonian villages [
13], we reasoned that random sampling in each village would result in a sample demographic similar to the population demographic. All villages in Djohong District were considered for possible selection. The World Health Organization’s (WHO) Extended Program on Immunization (EPI) method [
22] was used in the second stage to randomly select the 15 respondent households in each village. A household was defined as a group of individuals living under the same roof and eating meals from the same pot. Village chiefs assisted the research team to identify the geographic center of each village selected. From that point, the team randomly selected a direction by spinning a pen on a flat surface and then randomly selected a number of houses to pass to reach the first sampled household. Each subsequent household whose door was nearest to the door of the previous household was surveyed until all 15 surveys within the cluster were completed. To minimize non-response, a pre-visit announcement was sent to each village cluster to request the presence of all adult women in the village for the day of sampling. Three attempts were made to contact selected households where respondents were initially unavailable. When, as a result of PPS sampling, larger villages contained more than one cluster, clusters were geographically distributed according to the location of population centers.
Survey
At each village cluster, the study was announced as a women’s health study for purposes of safety and confidentiality. The team queried the adult female (≥18 years of age) head of household on household demographics, human security, household economy and assets, level of education, food security, water, fuel, shelter, access to health care, self-reported mental health, self-reported reproductive health and sexual violence experienced by the respondent only. In polygamous households, the senior wife was interviewed, a decision based on cultural custom.
For the purposes of this study, sexual violence was defined as any physically or verbally forced sexual act, including molestation, forced undressing, forced or unwelcome touching of a sexual nature, forced intercourse, forced insertion of an object into any body cavity or any other non-consensual sexual act, whether completed or not.
Perpetrators were defined as the person or persons responsible for the forced act. Refugees were defined according to the UNHCR definition: ‘owing to a well-founded fear of being persecuted for reasons of race, religion, nationality, membership of a particular social group or political opinion, is outside the country of his nationality, and is unable to, or owing to such fear, is unwilling to avail himself of the protection of that country [
23]’. Given considerations of confidentiality, no identification or proof was used to verify refugee status.
The survey instrument, much of which had been field tested the previous year for a baseline prevalence study on sexual violence, was translated and back-translated by bilingual French-English speakers in Cameroon, then tested and colloquially adjusted by data collectors who were local health professionals (with the Cameroonian Ministry of Health at Djohong Hospital) as well as local community health workers on staff with an international health non-governmental organization (NGO). Training included detailed explorations of each question with fine-tuning of terms and translation in French and in the local language, Fulfulde. Data collectors were then trained on the sampling methodology, including the EPI method, using simulated models of villages with varying configurations. Following training on the survey instrument, interviews were conducted in Fulfulde in a setting that ensured privacy and confidentiality.
Human subjects protection
This study was conducted in accordance with the WHO ethical and safety recommendations for researching, documenting and monitoring sexual violence in emergencies [
24]. Due to high rates of illiteracy, the team obtained oral consent prior to administration of the survey and specifically informed the respondents that the survey would ask about sexual violence as a women’s health and livelihood concern. Prior to survey questions on sexual violence, a second verbal consent was obtained to allow the participant an opportunity to refuse to answer questions in this section. Respondents were assured that their names would not be recorded, that there would be no penalties or benefits for refusing or agreeing to participate, and were offered access to counseling and medical services through the international NGO and local providers as needed. The study was reviewed and approved by the Office of Human Research Administration at the Harvard School of Public Health. All male data collectors were required to have previous experience in the care of female survivors of sexual violence, either as clinicians or as sexual violence counselors. While WHO guidelines recommend female surveyors and translators whenever possible, the lack of female staff and the availability of experienced sexual violence male counselors in the region necessitated their inclusion as surveyors. The survey instrument did not identify respondents.
Analysis
The lead field investigators checked the data for errors, then coded and entered it daily into a password protected Excel spreadsheet. Non-identifiable hard copies were stored in a locked facility on the NGO compound in Djohong District. At the end of the survey, the lead investigators securely transported hard copies to a locked storage facility at the Harvard Humanitarian Initiative where they remained accessible only to the lead investigators. STATA 11 (StataCorp, College Station, TX) was used for the analysis of the imported spreadsheet. Cluster sampling design was accounted for in the analysis; a generalized Hansen-Hurwitz estimator for a two-stage cluster design was used to estimate the means and percentages and confidence intervals were constructed by calculating the standard error of the generalized Hansen-Hurwitz estimator. Receiver operating curve (ROC) analysis was used to evaluate the predictive value of human security indicators for both lifetime and six-month sexual violence. Three predictive models were analyzed for their ability to predict sexual violence within the refugee and host populations: 1) age and ethnicity; 2) age, ethnicity, time in village, and all human security indicators; and 3) age, ethnicity, time in village and community human security indicators only.
Discussion
Refugees and host population report similar levels of human security in terms of attachment to home, community, and a positive grasp of the future. Though some human security indicators differ between these populations, overall they reflect a relatively stable population at the time of this study, where refugees not only feel safe but also have access to land, livelihoods, clean water, shelter, and wish to remain in their newly adopted villages for many years to come. Refugees appeared to be able to attain stability and security in Eastern Cameroon—in fact, though the proportion of refugees that owned land was lower than the proportion of host population that owned land, refugees and Cameroonians who owned land owned in similar amounts, suggesting equal opportunity to grow livelihoods between the two groups. Similarly, refugees had equal levels of assets as many host population households. Refugees and Cameroonians were both similarly attached to their villages, desiring to stay in the same village well into the future.
However, despite the unusual and encouraging stability and egalitarian quality of these human security findings, these human security indicators missed an epidemic of sexual violence that endangered the lives and health of refugee and host population women. ROC analysis shows that human security indicators measured in this study did not uncover either lifetime or six-month sexual violence. These data suggest that current, gender-blind means of describing human security are missing serious threats to the safety of one half of the population. If human security is ‘concerned with the protection of people from critical and life-threatening dangers’ [
25], this presents a major gap.
Why is it crucial that measures of human security be sensitive to gender? A robust body of evidence has explicitly linked measures of gender inequity and violence against women, including intimate partner and domestic violence, to higher levels of conflict [
8,
12,
26‐
29]. Much of this work has been done by Caprioli, whose analyses have shown that increased domestic gender equity has a pacifying effect on state behavior at an international level [
26], and that states with higher levels of gender equality are less likely to resort to violence first in resolving international disputes [
27]. Caprioli went on in 2005 to link lower levels of gender equity to higher rates of internal conflict [
8], and more violent conflict [
28]. Melander reproduced these findings, demonstrating that societies with higher female representation in parliament and higher female to male education attainment ratios have lower levels of internal conflict [
12]. There are several theories as to why this may be true. Some suggest that a tendency toward peaceful behavior is associated with a tolerance of the rights of others [
29], while others cite that states with a tendency toward inequity and oppression are inherently likely to be more violent both internally and internationally [
8,
11]. Violence against women not only has implications for the human security of half of the world’s population, it has consequences for the stability of the state.
Human security indicators represent current feelings of respondents, and thus are not likely to be sensitive to distant events. Lifetime sexual violence could have occurred any time over the respondent’s lifetime, and in either CAR or Cameroon, making it difficult to interpret the meaning of high human security indicators in the context of a high prevalence of lifetime sexual violence. However, human security indicators were also not associated with any meaningful association with sexual violence during the past six months, suggesting an insensitivity of this model to a major threat to the security and health of women and girls. Further, lifetime and six month prevalence for sexual violence reported by both women in the CAR and Cameroon were similar, and both proportions were substantial. Sexual violence represents one extreme manifestation of gender inequity—this level of violence is likely to be associated with other manifestations, including limited access to education and access to justice. Low levels of literacy, low rates of completion of primary and secondary school education, poor access to justice for survivors of sexual violence, and high rates of early marriage were outlined in a previous publication focused on sexual violence findings of this study [
17]. Thus, sexual violence against women represents a proxy indicator—it suggests that gender inequity has real human security consequences in this population, consequences that are not captured by this quantitative model, and consequences that extend beyond sexual violence itself.
These data also confirm what was described in the report of the Human Security Research Group published in 2012 [
30]. This report outlined data suggesting that rape as a weapon of war, while important in many contexts, does not constitute the majority of rape faced by women in conflict. The findings of this report support this assertion [Tables
8 and
9], as the majority of women who experienced sexual violence reported that the perpetrator was their intimate partner (64.0%, 95% CI 54.3-72.5) and the second most common perpetrator was a friend or member of the community (20.0%, 95% CI 14.2-27.5). The group identified as ‘combatants’ - soldiers, rebels, and
coupers de route— who, in most conflict contexts, would be the focus of sexual violence related insecurity—constituted a comparatively less commonly reported perpetrator (17.1%, 95% CI 10.7-26.1). When disaggregated data is examined [Tables
8 and
9], data show that while refugees report a higher proportion of sexual violence at the hands of combatants (39%, 95% CI 25.6-54.2) than Cameroonian women, refugees still report that a majority of the perpetrators of sexual violence are known to them (intimate partners: 52%, 95% CI 35.4-68.1: friends/community members: 15.6%, 95% CI 8.3-27.2). Based on these findings, we estimate 22.3% (95% CI 17.1-28.5) of all women in Djohong District, Cameroon have endured sexual violence by their intimate partner during their lifetimes. Several authors have reported high rates of sexual violence in times of conflict [
31‐
34], and as suggested by the Human Security Research Group, much of that occurs at the hands of intimate partners and community members. While a focus on sexual violence perpetrated by armed forces remains important, the much more common epidemic of violence perpetrated by intimate partners and community members against women must also be addressed, particularly given that this violence presents risks not only for the affected women but for the development and maintenance of their communities and the world as a whole.
Several measures have been developed to understand gender equity at the country level, including UN Women’s Indicators and Statistics Database, GenderStats, and the World Economic Forum’s Gender Gap Project, and many studies analyze other measures including female to male education attainment ratio, female representation in parliament, life expectancy, literacy, and participation in democracy. However as Hudson points out, these do not include measures of physical violence against women [
29]. These are also aggregated sources at the country level, and not likely to aid in a nuanced understanding of community level human security, as is the goal of the Leaning and Arie model.
Based on the findings of this study, we suggest that, at a minimum, measures of human security should include prevalence of sexual violence at the hands of all types of perpetrators, including both combatants and non-combatants. Additionally, measures of gender inequality that have been previously linked to conflict in other research (female participation in parliament and female to male higher education attainment ratio) should be included in human security frameworks.
Limitations
All interviews were conducted privately in their homes in the respondent’s preferred language. Nuanced understanding of questions at the household level could be distorted in the course of translation and risked being asked differently in Fulfilde across data collectors. Though all efforts were made to ensure the respondents privacy at the time of the interview, it is possible that women felt reluctant to report sexual violence given strong social stigma associated with sexual violence. Whenever possible, respondents were interviewed by women, however, several members of the surveying team were male, due to limitations in availability of experienced multilingual staff. Although the few male data collectors had prior experience in medical and psychological intake of sexual violence survivors, it is possible that women felt less comfortable reporting their experiences with sexual violence to them regardless. We proceeded with male data collectors based on the acceptance in this community of male sexual violence counselors in a pre-existing international NGO program on sexual violence. Although female data collectors are recommended according to WHO guidelines [
24], several studies have successfully engaged male data collectors in similar research [
19,
33]. As all data reported here refer to the respondent’s last episode of sexual violence, it is likely that these statistics under-represent the true burden of sexual violence in this community. Results are specific to refugee and host population female heads of household over the age of 18 only, thus results cannot be generalized to all women in the region and are likely biased towards the experiences of older women. Given these female heads of household tend to be first wives in polygamous households, this study represents the experiences of this group of women primarily. Arguably any form of sexual violence could affect human security and data was not collected on men or children.
Several of the data collectors were NGO sexual violence program staff and thus known to be affiliated with these programs. Although interviewers were carefully trained to emphasize that no aid or compensation would be given for participation in the survey, it is possible respondents may have altered their responses. Respondents may either have reported themselves to be refugees in the interests of obtaining services, or Cameroonians in order to avoid stigma. Given the anonymous nature of this survey and initial disclosure regarding aid and compensation, we do not feel it is likely that this issue significantly impacted responses.
Human security indicators were developed by the research team based on the Leaning-Arie model, used during a 2009 iteration of a population based survey in the region and refined based on the results. Outcomes are self-reported, thus assets, income, and land holdings were not verified objectively.
Competing interests
The authors have no competing interests to declare.
Authors’ contributions
PKP was involved in study design, data acquisition and analysis of data, and primarily drafted the manuscript. PA was involved in field data acquisition, analysis of data, and manuscript preparation. JS was involved in analysis of data and manuscript preparation. RG directed statistical analysis of data and was involved in manuscript preparation. PGG was involved in study design, analysis of data, and manuscript preparation. All authors read and approved the final manuscript.