Study setting and participants

The study was conducted in three border towns located in three major human trafficking corridors in Ethiopia, namely Mettema-Yohannes, Moyale and Galafi from May to October 2016. Mettemma-Yohannes, Moyale and Galafi towns are geographically located bordering Sudan, Kenya and Djibouti, respectively. These study sites are situated on the most important gates through which many Ethiopians are being trafficked to major destination countries. The study sites may be helpful to get a representative sample to study the trafficking experiences of victims in the country. The study participants were Ethiopian women who were trafficked mainly to the Middle East, South Africa and Europe or were in transit countries and came back either by deportation or voluntary return. They were recruited at the waiting rooms of immigration offices in the border towns to which returnees were expected to report or in hotels they booked in those towns.

The trafficking status of returnees was ascertained by the UN 2000 definition of human trafficking. Thus, if a returnee was either younger than 18 years at the time of departure or recruited by deception, fraud, cheating or coercion with subsequent exploitation, then she would be considered as trafficked. Exploitation could be apparent in several forms, like working for extra hours per day, not paying their wage, transferring and selling them to traffickers or smugglers. The detail of the definition we implemented was presented in the other part of this work, and it was found that about half of the returnees were trafficked.22

Study design

A retrospective cohort study was designed to examine sexual violence among female returnees. Data related to sexual violence experienced at travelling, destination and detention stages were collected separately and retrospectively. It was performed by encouraging participants to recall events of sexual violence within those stages of trafficking. A pilot study conducted on 196 returnees had already informed us that they could recall episodes of violence they experienced up to the past 2 years. We also understood that returnees who were trafficked for at least 3 months were exposed to most of the stages of the human trafficking process. Thus, returnees who were in trafficking condition for the preceding 3–24 months were recruited and gave the necessary data from their experience. However, only a few migrants were excluded from the study due to this criterion.

Sample size and sampling procedure

The sample size of this particular study was 671 women and girls who participated in our other larger study that involved both male and female participants. In the larger study, there were two stages of sampling. The primary sampling unites were border towns situated in the major trafficking corridors in the country, and the secondary sampling unites were victims of trafficking who were returning home via those border towns. Three border towns were selected with simple random sampling method out of four border towns situated in the major trafficking corridors in the country. As a secondary sampling unit, all Ethiopian returnees via the three border towns who met the UN 2000 definition of human trafficking were recruited consecutively and concurrently until a sufficient sample size was secured starting from May 2016. Of all participants approached in the larger study, about 90.95% of them complied with it giving a total of 1387 male and female participants.23 Thus, all the 671 women and girls who were a part of the larger study were taken as study participants for the current study.

The adequacy of the sample size in estimating the proportion24 of sexual violence among female returnees was checked by assuming a prevalence of sexual violence as 50%, a confidence level of 95% and a design effect of 1.5 (because we had two stage sampling: the trafficking corridors and participants) that gave us 577 trafficked women. To increase the precision of the estimate, we included all the available 671 female returnees.

Data collection tools and procedures

A structured questionnaire was developed after reviewing available literature1 6 13 14 18 25 26 and supplementing it with contextual understandings of study settings and participants from our qualitative study.27 Four data collectors were assigned to Metemma-Yohannes and two each to Moyale and Galafi towns. The data collectors, half of whom were women, were given training for 2 days. The training also included how to approach victims of trafficking and how to ask them sensitive questions. Each victim of trafficking was interviewed separately by a data collector of the same sex. The data collectors chose suitable places to conduct the interviews where participants would be encouraged to respond freely, including near the waiting rooms of each immigration office or the hotels they booked. Thus, returnees were contacted only once and gave their experiences retrospectively. In situations where respondents and interviewers could not speak a common language, interpreters were employed. In addition to all these efforts made to maintain the validity and reliability of the tool, the principal investigator and field supervisors closely supervised the data collection at the time of the survey. No incentive was given for participation.

Variables and measurement

The outcome variable of the current study is sexual violence. Items used to determine the status of sexual violence were based on the WHO international study of domestic violence.28 Thus, the three items of sexual acts include forced to have sex, had sexual intercourse because of fear if she was not compliant to it or being forced to do something sexual that she found humiliating. One or two of the items in sexual violence assessment tool were very sensitive. Therefore, after reading all the three items to each respondent, we asked whether she had experienced at least one of them; a positive response meant the respondent was a victim of sexual violence.

The status of sexual violence, labelled as ‘yes’ or ‘no’, was determined for each woman or girl at each stage of the trafficking process: predeparture, travelling, destination, detention and integration/reintegration. However, because all returnees were interviewed when they were just entering Ethiopia from abroad, data related to sexual violence experienced during integration/reintegration was not found to be complete and thus not considered in any analysis. However, predeparture stage mainly included the time before the start of the actual trafficking process. Therefore, sexual violence at predeparture stage was not considered as an outcome variable during the regression analyses; instead, it was analysed only descriptively.

The potential predictor variables for sexual violence were sociodemographic characteristics (age, marital status, educational level, residence, social support, type of job at destination and family wealth quintile), trafficking-related exposure variables (restricted freedom of movement, smuggling status at departure or initially smuggled, trafficking corridor, history of detention by foreign security, time spent at each stage of trafficking and postdeparture physical violence) and behavioural and other characteristics (predeparture violence, predeparture substance use, and postdeparture substance use). There are also other variables that were used for descriptive purpose.

Age was grouped in a very similar way to those of two studies conducted across seven European countries where the study participants came from most continents.12 18 The time spent at each stage of the trafficking process was measured in weeks and considered as a continuous variable; other variables such as marital status,12 18 29 educational status,6 30 residence, region and religion30 were treated by referring to literature or considering the context of the current study setting or participants. Moyale town, which was one of the labels of the variable named as ‘Exit corridor’, was found to have a very small number of females. Therefore, it was merged with other exit gates and labelled as ‘Others’.

To assess participants’ perception about the level of care they were receiving from others before departure, the Oslo social support scale with three items was employed.31 The first question has four alternative responses, while the second and the third items each has five options. The sum of scores for each participant would range from 3 to 14. A sum score between 3 and 8 was classified as ‘poor support’, 9 and 11 as ‘intermediate support’ and 12 and 14 as ‘strong support’.

A composite wealth index was determined at household level using principal component analysis. Indexes generated from urban and rural specific items were weighted to determine the composite wealth index (online supplementary file). Three of the 11 items in the scale ask about household size, whether all children aged 6–12 years were attending school, and construction material of the walls of dwelling unit. It was adopted from the Ethiopian demographic and health survey.32 Moreover, the presence of ‘restricted freedom of movement’ was checked by interviewing whether each participant had been locked in a house or a building or never be free to go anywhere she wanted or do what she wanted by warning or terrifying her.6 A returnee would be considered as ‘smuggled at departure’ (or ‘initially smuggled’) if she took a contractual agreement with a person that would take her illegally starting from Ethiopian border to any other destination. To minimise the temporal dilemma with human trafficking, a smuggling process that was started in other transit countries was not considered as positive for ‘initially smuggled’ as far as she crossed legally the border between Ethiopia and a neighbouring country.

Statistical analysis

Descriptive statistics were reported using tables. In determining factors associated with sexual violence among women and girls, a logit family generalised linear mixed effect model was employed to account for the dependency of repeated measurements at each stage of trafficking. The models included both the fixed effect of each covariate and the random effects associated with each participant from whom repeated data were taken. The first mixed model considered the fixed effects and random intercepts. However, it was not better in fit than the ordinary logistic regression model when tested using the likelihood ratio test (p value=0.913). The second model that added a random slope on the first mixed model did not have any significant gain in fit when compared with either of the two models using the same test. As a result, emphasis was given only to fixed effect interpretations for a model containing both the fixed effects and random intercept.

The association between each variable and sexual violence was measured by OR, and when it was controlled for other variables, adjusted ORs (AOR) was reported. We conducted univariable analyses containing both a fixed effect for each independent variable and random intercepts. Independent variables with a p value of <0.20 were carried forward for inclusion in the multivariable analysis.33 However, in the multivariable analysis, only independent variables with p values of 0.05 or less were considered statistically significant.

During the analyses, we assumed that the missing mechanism of data was non-informative; that is, it is unlikely to get systematic differences in regards of sexual violence and other key variables between those who agreed to and declined to participate. The rationale behind this assumption was related to circumstances prevailed at the time of data collection that persuade returnees not to participate in the study. There was no reason to assume that the missingness of data was related to the values of observed or unobserved study variables including the outcome variable. Few of the reasons for non-response that encountered during the interview were the following: sometimes, returnees may go to their origin without being interviewed simply because of some personal reasons such as when interviewers were not available on time to collect data at the immigration offices; occasionally a large number of returnees arrive at the immigration offices, and as a result, it could be difficult to interview all of them within a time period before they leave to go home; and sometimes, respondents may not speak the language of interviewers or translators assigned and thus may not be involved in the study.

Patient and public involvement

This study did not involve patients and the public.

Ethics and confidentiality

We followed the WHO ethics protocol recommended to interview trafficked women and attempted to ensure that all the field workers were sensitive to the needs, privacy and fears of participants and not to retraumatise them during the interviews.34 Ethics approval statment was obtained from the Institutional Review Board of the University of Gondar. We also first obtained permission from Ethiopian Immigration Office Headquarter and then from all the three Immigration centres located in the three border towns that were selected for the study. A written consent form describing the aims, benefits and potential risks of participation in the study as well as that containing information about their right to withdraw was given or read for each adult participant, and the willingness of each returnee was asked before they sign it. For children whose age were 17 years or less, assent form was prepared and consent was obtained from the organisation that took care of them. We also implemented the International Organisation for Migration data protection principles to maintain its confidentiality and anonymity.35

Background characteristics

The study included 671 female returnees of human trafficking. The mean (SD) age of the victims at the time of departure was 20.94 (3.48), and the youngest and oldest ages were 14 and 49 years, respectively. Participants were in trafficking conditions for 3–24 months with a mean (SD) of 15.37 (5.76) months. Of all participants, 61 (9.1%) were either illiterate and/or attended informal (eg, religious) education only, 236 (35.2%) completed elementary school and 202 (30.1%) completed high school or above (table 1). The majority or 480 (71.5%) of the participants returned through Metemma-Yohannes, while 180 (26.8%), 2(0.3%) and 9 (1.3%) came via Galafi, Moyale and other gates, respectively.

Sexual violence among trafficked females

Over two-third (prevalence=71.1%; 95% CI 67.5 to 74.4) of female victims of human trafficking were sexually violated during the whole trafficking period. The prevalence of sexual violence was 10% (95% CI 7.9 to 12.5) during predeparture, 35.0% (95% CI 31.5 to 38.7) travelling period, 58.1% (95% CI 54.2 to 61.8) at destination and 19.5% (95% CI 15.2 to 24.6) at detention stage. Considering the time spent at each stages of trafficking of women and girls, ‘travel stage’ had a mean (SD) of about 2.22 (1.35), ‘destination’ 56.73 (24.04) and ‘detention’ 1.23 (1.95) weeks (table 2).

Regarding the incidence rate of sexual violence during the whole trafficking period, it was estimated that about 11.81 females experienced the events of sexual violence per 1000 person-weeks. During the travel, destination and detention stages, about 157.64, 9.90 and 65.62 women experienced the event of sexual violence per 1000 person-weeks, respectively.

Sexual violence by various characteristics along the stages of trafficking

Among female returnees who were teenagers at the time of departure, 48 (64.9%) were sexually violated during the whole trafficking period. Considering the travel stage characteristics, 331 (69.0%) of the female victims of trafficking who left their origin via Metemma Yohannes and 123 (68.3) through Galafi corridor were sexually violated while they were in the trafficking situation. For the characteristic named as ‘initially smuggled’, 372 (70.9%) of the initially smuggled and 91 (62.3%) of the initially non-smuggled females were sexually violated during the whole trafficking period (table 3).

Risk factors of sexual violence

Of the 16 potential predictor variables, seven of them were excluded from the multivariable analysis because they were insignificant contributors in the univariable analysis as their p values were more than 0.20. Therefore, the remaining nine variables were examined in the multivariable analysis out of which four variables, namely marital status, residence before departure, exit corridor and history of detention abroad, did not contribute significantly to the model. Thus, the odds of experiencing sexual violence for teenagers aged 14–17 years was two times higher than that for the oldest groups aged 26–49 years (AOR=1.97, 95% CI 1.11 to 3.52). Similarly, women in the 18–20 and 21–25 age groups had each almost twofold odds of experiencing sexual violence than women aged 26–49 years.

After adjusting for other variables, the odds of being sexually violated for initially smuggled women was about one and half times higher than for their non-smuggled counterparts (AOR=1.45, 95% CI 1.09 to 1.93). Similarly, postdeparture physical violence (AOR=2.94, 95% CI 2.24 to 3.86) and restricted freedom of movement (AOR=1.45; 95% CI 1.13 to 1.86) were positively associated with sexual violence. Time spent at each of the three middle stages of human trafficking, namely travelling, destination and detention, were positively associated with the odds of sexual violence (table 4).

Limitations of the study

The current study could represent a wide range of the population of trafficked persons who were under various forms of exploitation. However, the extent to which it could represent victims who would not return home for various reasons is unclear. Moreover, though both female and male victims are vulnerable to sexual violence during trafficking, this work couldn’t involve males in the sexual violence study because of the unreliable data obtained from this group in the pilot study.  Due to the sensitive nature of sexual violence, even women and girls might under-report it, especially forced sexual events. Of course, though women and girls might under-report the severe forms of sexual violence, like rape, they usually report them in less severe terms that are included in the definition of violence.  

The number of female returnees who were trafficked for commercial sexual exploitation were very few, and this is unclear whether the occupation was under-reported due to its sensitive nature or not. On top of that, because the design we employed was retrospective, under-reporting due to recall bias might not be ruled out. Of course, some of the events and experiences they encounter during the difficult times could minimise recall bias as they are ever unforgettable. Besides, we made thorough efforts to help women and girls remember their experiences.