In 2018, PHR conducted a retrospective study with Rohingya refugees in Bangladesh to capture the experiences of people from all Rohingya hamlets in northern Rakhine state whose populations had been displaced to Bangladesh. The study had three arms: (1) a quantitative hamlet-level survey, (2) qualitative in-depth interviews with hamlet leaders, and (3) clinical evaluations of survivors to document physical sequelae of violence. Each of these arms have been documented in prior publications and are summarized below [11,12,13].
The study was conducted in the Rohingya refugee camps in Ukhiya and Teknaf upazillas in Cox’s Bazar District, Bangladesh. All of the respondents had arrived from Myanmar by October 2017, and all had previously resided in northern Rakhine state.
An overview of the administrative divisions in Myanmar is necessary to understand the research methods. The majority of Rohingya who lived in Myanmar prior to August 2017 resided in Maungdaw District, which, along with parts of neighboring Sittwe District, is generally referred to as northern Rakhine state (Fig. 1). Within this District, there are three townships: Rathedaung, Buthidaung, and Maungdaw. Each township has between 80 and 100 village tracts, which are each comprised of a group of villages, known locally in English as hamlets. A visual depiction of these administrative divisions is below in Fig. 2.
In the quantitative arm, hamlet-level data was collected between May and July 2018 from 604 leaders of hamlets and urban wards in northern Rakhine state (Maungdaw, n = 346; Buthidaung, n = 229; Rathedaung, n = 29) . In the Myanmar administrative system, hamlet and ward leaders are responsible for the regular reporting of local population data to the Myanmar government. These same individuals assumed leadership roles during the displacement to Bangladesh and continue to represent their former communities in the refugee camps. Based on this and through extensive consultation with the Rohingya refugee community and camp leadership, these leaders were determined to be representative interviewees for reporting on what happened to villagers before, during, and after flight to Bangladesh.
The surveys and in-depth interviews were conducted privately by trained Rohingya refugees representing each of the three affected townships from northern Rakhine state. Consenting participants were asked to report the frequency, scale, and type of violence experienced by the residents of their respective hamlets/wards during the offensive of August 2017 and their subsequent journey to Bangladesh . If respondents reported a certain threshold of violence in their hamlet (10 or more deaths of villagers, mass rape, and/or witnessing of mass graves in their hamlet or during displacement), the respondent was invited to participate in an in-depth interview.
One third of quantitative survey respondents qualified for in-depth qualitative interviews. This was a higher number than anticipated, and due to resource constraints (e.g., time and money), qualitative interviews ended early. Interviews were conducted with 45% of those hamlet leaders who qualified, for a total of 88, with respondents from hamlets across all three townships of northern Rakhine state (Buthidaung, n = 42; Maungdaw, n = 34; Rathedaung, n = 12) . The semi-structured interviews documented first-person testimonies of the violence and destruction that occurred within each respondents’ community, including attack details, reasons for fleeing, and experiences living in the settlement camps, among others .
The clinical evaluation arm documented physical injuries sustained by Rohingya survivors. Respondents were selected using purposive and snowball sampling to identify candidate participants who had experienced violence as evidenced by their physical scars or disabilities. Evaluations were completed among 101 individuals from 31 hamlets (Buthidaung, n = 7; Maungdaw, n = 19; Rathedaung, n = 5) by physicians trained in clinical evaluations during several visits from December 2017 through July 2018 . Individual interviews were conducted during the evaluation to corroborate examination findings with geographic location and type of violence. Data from three individuals were unable to be location-matched based on available location data .
Internal validation was conducted if data existed for the same hamlet from at least two study arms (i.e., quantitative, qualitative, or clinical). Key indicators of interest were reviewed for comparison, including: arson, presence of mass graves, sexual assault, injuries, and fatalities. Given that measures varied in their structure across each arm (for example, closed-ended measures in the quantitative survey vs. open-ended text for in-depth interviews), a binary classification (yes/no) was created of whether each of the above indicators was reported for a specific hamlet or ward in each study arm.
Analyses were then conducted to assess whether indicators across the three study arms were consistent. Consistency was defined as the presence of a positive indicator across two or more internal data sources. While the quantitative survey and qualitative interview arms collected hamlet-level event data, the clinical interviews collected individual-level data about experiences of violence and physical and mental health outcomes. Thus, events included in those clinical interviews should generally be captured in the quantitative and qualitative data, but events reflected in the quantitative and qualitative data sources might not be documented in any particular individual narrative. The quantitative survey data is the most comprehensive, as respondents were prompted to select a response from pre-specified, closed-ended options for each of the indicators for comparison. Conversely, the qualitative interview was semi-structured, and the clinical evaluation was unstructured. Thus, a negative indicator in the qualitative or clinical data was not defined as inconsistent with a positive indicator in the quantitative data; however, a negative indicator in the quantitative data that was positive in the qualitative data or clinical evaluation is considered contradictory. The inclusion of three methods of data collection resulted in a broad base of comparable data, with more nuanced detailed narratives at both the hamlet and individual levels. Since the quantitative study arm was most comprehensive and had the largest sample (N), it served as the control method for comparison; it was equally as important to perform qualitative and clinical measures to capture data that may have been missed by the yes/no format of the quantitative survey.
External validation and triangulation
External validation and triangulation were also conducted with data from other publicly available data covering the 2017 violence in Rakhine state. A literature review was conducted to identify and extract comparable data. This approach was modeled after the PRISMA framework (Fig. 3).
An initial group of documents was collected from human rights organizations and other trusted organizations known to have reported on the violence against the Rohingya. These sources were chosen for their recognition as foremost human rights organizations, with widely respected and ethically sound reporting methods. The majority of the external data represented in this report is from one of nine main sources. The types of data collected varied across sources. Qualitative interviews were conducted by Amnesty International, Fortify Rights, Human Rights Watch, and Médecins Sans Frontières. Additional survivor testimony was reported by the Associated Press, BBC News, Reuters, and the UN’s Independent International Fact-Finding Mission on Myanmar (FFM). Extensive collections of satellite imagery and other remote sensing data were published by Amnesty International, Human Rights Watch, Reuters, the Women’s Refugee Commission, and the UN’s FFM. Several of the organizations utilized a mixed-methods approach to document these human rights abuses. For example, Amnesty International used NASA satellite imagery and environmental sensor data to detect evidence of fires in Rohingya hamlets across northern Rakhine state . Additionally, it conducted both clinical examinations and qualitative interviews .
The websites of these organizations were reviewed for relevant publications and searched using key terms, including “Rohingya,” “Rakhine,” and “August 2017”. In total, 312 publications documenting the 2017 violence in northern Rakhine state were identified from these sources.
Additional literature sources were identified using the search engines Pub Med, Lexis Nexis, and Google. Key word searches included the terms “Rohingya” and “August 2017” with or without “Myanmar.” These searches yielded an additional 116 publications. Sources were also identified through citation mining, which contributed five further publications. Altogether, 426 unique records were identified, including articles, testimonies, comprehensive reports, qualitative interviews, quantitative health surveys, narrative journalism, and satellite imagery and other remote sensing data.
An initial rapid screening of titles and abstracts for relevance excluded 235 records, leaving 191 articles whose full text was assessed for eligibility (see Fig. 3). Literature was assessed for inclusion using the following criteria: accounts of violence perpetrated in the second half of 2017 in northern Rakhine state against the Rohingya; findings reported from primary data collection; published in English; data was available at the hamlet or village tract level; document was published and publicly available. The criterion by which sources were most frequently excluded was ‘failure to include primary data’.
Once relevant external literature sources were identified and extracted for triangulation, they were reviewed for village tract- or hamlet-level data that reported the following five key indicators: (1) arson, (2) presence of mass graves, (3) sexual violence, (4) human injuries, and (5) human fatalities. Each instance of violence that was reported in a specific administrative division was recorded. Qualitative interviews, testimonies, videos, and quantitative data were considered for each indicator. Satellite images and other remote sensing data provided information on arson. In all cases, the presence or absence of an indicator in a given source was assigned a binary (yes/no) designation for that indicator.
Standardization and alignment of locale names were required before data could be compared across sources. Due to the absence of an official list of hamlet and village tract names in both Burmese (the official state name), and Rohingya (a locally used name), a customized list was generated based on the Myanmar Information Management Unit (MIMU), input from non-governmental organizations which had operated in Rakhine state, and input from Rohingya research staff. Transliteration of Burmese and Rohingya names into English followed best practices established by local researchers and referenced either the MIMU or predominant spellings in existing Rohingya literature.
Rohingya is an oral language that does not have a written form. Because there is no standardized method for transliterating either Burmese or Rohingya words into English or Burmese script, this often results in an array of spellings for the same word. As a result, the spelling of geographic location names (i.e., township, village tract, hamlet) varied based on how they were recorded by individuals ranging from trained Rohingya data collectors to temporarily resident foreign journalists. This was especially true for data collected by different organizations, but even within the PHR data, there was sometimes variation between each of the three study arms. For example, “Pa Da Kah Ywa Thit” (clinical), “Pada Kar Ywa Thit” (quantitative), and “Ba Da Ga Ywa Thit” (qualitative) are different transliterations for the same hamlet. For Burmese names, the English spelling used was that designated by MIMU. For the Rohingya names, English spelling was standardized based on input from Rohingya researchers and generally following the predominant spelling in the literature.
Some locations were incorrectly identified in the open-source literature. In some cases, village tract names were presented as hamlet names, hamlet names that occur in multiple village tracts were given without specifying to which village tract they were referring, or the hamlet named was not part of the village tract to which that document assigned it. If a location was unclear due to mismatched hamlet and village tract information or alternative spellings, the location information was confirmed with the source organization, if possible. Most locales were readily identified, with only eight locales—representing 2% of the 348 locations specific accounts identified—for which a match could not be found. These eight locales, from five publications, were excluded from the comparative analysis, as there was no comparison data.
Data was then aggregated at the hamlet level. For information that identified the village tract but not the hamlet, data was aggregated at the village tract level. The aggregated data for an indicator of interest were presented as a fraction, with the numerator representing sources positive for that indicator (i.e., “yes” or present), and the denominator representing the number of sources potentially reporting on that indicator. For example, if five of the six sources reporting violence in Ah Htet Nan Yar hamlet identified the presence of arson, the arson indicator for Ah Htet Nan Yar would read 5/6.
Because most of the reports by other organizations presented only partial accounts or quotes from respondents and were not intended to be comprehensive accounts of occurrences in a particular hamlet, the PHR data was considered the baseline for comparison (e.g., internal data). External data was considered consistent with internal data if the internal indicator was positive and any of the external sources were positive, or if the internal indicator was negative and all of the external sources were negative. It is important to note that while a positive indicator signifies that the event in question was reported at that location, a negative indicator means only that it was not reported, rather than that it definitively did not occur.
Following aggregation, external data was assessed for consistency by locale and by indicator. For each locale, external data was compared to internal data for each of the five key indicators. Findings for a particular location were considered to be consistent overall if greater than 50% of the indicators were consistent (e.g., three or more of the five key indicators). Consistency was also assessed by indicator, across all locations, by calculating the proportion of hamlets for which sources reported consistently for that indicator.
Results are presented only in aggregate, without identifying the findings of each source individually. This is to protect the confidentiality of survey and interview participants who were hamlet leaders and thus identifiable by the hamlet name . For that reason, PHR has not previously published findings at the hamlet level, and herein hamlet-level data is only presented in aggregate when there are at least two other sources. In order to ensure that we may continue to protect the identities of our subjects, the only hamlet-level data provided here is reporting consistency between study arms (Tables 1, 3, 4) and does not indicate whether or not the respondent reported instances of violence within a particular hamlet.
For two locations—Maung Nu and Min Gyi—where major events occurred and for which there are a high number of reports, more detailed analysis was conducted (Tables 3, 4), including numbers of injuries and fatalities, perpetrators, and types of violence. In these instances, we presented analysis at the hamlet level, but did not disclose which source reported which data so that a specific respondent’s reports could not be identified.
Ethical review and human subjects considerations
The PHR Ethics Review Board (ERB) provided full ethical approval for this study. Because no formal Rohingya body exists that could serve as a review board, PHR held a community consultation with Rohingya leadership before administration of the qualitative and quantitative components of this work, to obtain their input, feedback, and approval. PHR’s ERB reviewed and approved the clinical arm. All study participants underwent an informed consent process in Rohingya language before participating in any of the study arms.
In order to protect respondent confidentiality and protect them from possible recrimination from the Myanmar government, no identifying information was recorded beyond hamlet of origin and leadership role of each respondent. Data was collected on password-protected tablets and transferred for storage on secure password-protected and encrypted servers. The research team discussed risks of this survey and their corresponding mitigation plans with all respondents, the Rohingya data collection team, and Rohingya leadership during community consultations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.