EPO Sourcing Methodology

Each week, ACLED EPO researchers review dozens of English, Amharic, Somali, and Afaan Oromo language sources, in addition to information gathered from local partners, to code political violence and demonstration events in Ethiopia. Ethiopia’s information environment has gone through several shifts, and this document outlines how ACLED and the EPO have addressed these changes.

In contexts with highly dynamic information environments such as Ethiopia, ACLED regularly reviews its sources to ensure reliable information for the data set (for more, see this methodology primer). The conflict in Tigray that began in November 2020 highlighted that the need to constantly adapt to challenges unique to a specific context is especially important in Ethiopia. ACLED has made an effort to expand coverage of events around the country by consulting local media experts and in-country organizations. As a result, multiple new local sources have been added to ACLED’s source network. 

Additionally, ACLED has developed the EPO to further enhance data collection, cover conflicts throughout the country in greater depth, and produce more extensive analysis of key trends. The details of the EPO’s sourcing approach are outlined below. 

Challenges to sourcing in Ethiopia

Prior to 2018, ACLED’s greatest sourcing challenge in Ethiopia was overcoming the country’s tightly controlled media environment. The absence of free and independent media, the routine arrest of journalists and reporters, and repressive legislation curtailed the media’s ability to report from across the country, particularly in areas where armed insurgencies and government opposition have been strongest. In 2017, Reporters Without Borders ranked Ethiopia as one of the world’s most censored countries,1Reporters without Borders Index, 2018 underscoring why, prior to 2018, ACLED’s Ethiopia sourcing relied more heavily on diaspora-based sources and international media outlets. Through engagement with local sources, these outlets were able to report on the drastic increase in demonstration activity that preceded the power change in 2018. 

After Prime Minister Abiy Ahmed came to power in April 2018, the government began to open up Ethiopia’s media landscape.2James Jeffrey, ‘The challenges of navigating Ethiopia’s new media landscape,’ Al Jazeera, 29 October 2019 He removed bans on over 250 media outlets and freed dozens of jailed journalists.3Maggie Fick, ‘In Abiy’s Ethiopia, press freedom flourished then fear returned,’ Reuters, 28 May 2021 While his sweeping reforms resulted in the proliferation of media outlets and encouraged freedom of speech, several challenges remained, including media bias and government shutdowns of the internet in response to security threats.4Ethiopian Reporter, ‘Council urges authorities to lift internet restrictions,’ 4 March 2023; Abdur Rahman Alfa Shaban, ‘ Ethiopia ‘fully’ restores internet connectivity after three week outage,’ 23 July 2020; Elias Meseret and The Associated Press, ‘Ethiopia Is Coming Back Online After Government Assassinations Prompted a 10-Day Internet Blackout,’ Fortune, 2 July 2019; Fidelis Mbah, ‘Outrage over Ethiopia’s continuing internet blackout,’ Al Jazeera, 25 June 2019 

Political instability fueled by multiple insurgencies against Abiy’s government has challenged the prime minister’s liberalization agenda since 2019. Government pressure, increased polarization, and access constraints in conflict areas have resulted in a more restricted information environment, though restriction levels vary across the country. Since 2019, the number of distinct media voices and sources has decreased; journalists have been arrested, deported, and harassed, limiting their ability to provide coverage for both domestic and international audiences; media have increasingly reflected the polarized nature of their audiences; and bias in the coverage of individual events has increased.5Declan Walsh, ‘Ethiopia Expels New York Times Reporter,’ New York Times,’ 20 May 2021Committee to Protect Journalists, ‘Ethiopia expels Economist correspondent Tom Gardner,’ 16 May 2022 Meanwhile, some regions are ‘media deserts’ without accessible public channels to report on activity. Other regions, including those with active conflicts, have suffered from limited coverage due to media, communication, and electricity blackouts, as well as access constraints linked to security risks.6Associated Press, ‘Ethiopia offers no date for end to blackout in Tigray region,’ 29 November 2022 Thus, ACLED’s approach for sourcing information in Ethiopia has evolved several times since 2018, as information options multiplied as diaspora media returned to operate nationally, and new outlets emerged. Yet, simultaneously, sources of misinformation also proliferated alongside openly biased media outlets. 

Conflict environments are already ‘low information’ spaces, as the identity of perpetrators and victims and the intensity of violence and its outcomes are difficult to ascertain accurately in real time. Through social and traditional media, a flood of reports, accounts, and rumors can distort the reality of a conflict. Reporting standards for verified information can be subsumed in this cacophony of distortion. Reporting around Ethiopian conflicts are rife with propaganda, unsourced reports, reports without reliable sources, exaggerations, contradictory information, incomplete information, and unreported events. Each conflict across the country is affected by one or more of these problems: In several conflict-affected regions, information is either very sparse (e.g. in the Benshangul/Gumuz region), distorted and often based on rumor and propaganda (e.g. parts of Tigray and Amhara regions), subject to exaggeration (e.g. Oromo Liberation Army (OLA) actions in Oromia region), or incomplete (e.g. many parts of Tigray and the Afar-Somali regions). Compounding these issues, social media sites have increasingly played a role in pushing mis- and disinformation from well-established diaspora and external communities. 

Further, there are specific areas in which the Ethiopian government demands, or the persistence of insecurity effectively led to information blackouts. In these cases — such as in Tigray between November 2020 and November 2022, and in the Metekel zone in the Benshangul/Gumuz region after 23 January 2021 — there were ongoing accounts of human rights abuses, offensives, rural violence, and evidence of ‘ethnic cleansing’ that could neither be detailed nor confirmed by reputable sources until many months later. 

ACLED’s Rigorous Verification Approach

When heightened insecurity and instability lead to a decrease in reporting and information, ACLED engages with local and national partners, journalists, translators, local fixers, and stringers to determine the veracity and details of available claims before coding. If that is not possible, due to various reasons like communication blackouts, ACLED will account for the full series of events once news agencies or human rights organizations are able to verify the details of the events. In the meantime, ACLED will note the limitations of the data in the notes column for transparency or add a note on EPO reports. 

This same logic applies to individual events and those that are speculative: The component parts of each event must be available for an event to be included in the dataset and in larger analysis. This means that certain events — such as those reported in aggregate, or incidents without specific information about time and location — cannot be coded by ACLED’s researchers.7For more, see the Event Types and Sub-event Types section of ACLED’s Codebook. However, researchers keep records of unconfirmed accounts so as to integrate credible reports into the dataset when further sources, information, or confirmation becomes available. Finally, new information  is integrated into the dataset to reflect the most updated and accurate information about an event or the broader dynamics of a conflict, where relevant. In this way, ACLED is a ‘living dataset,’ with new and historical events regularly added and updated.8For more, see Is historical ACLED data ever revised? 

ACLED consistently seeks to engage with and depend on a network of local partners to ensure the accuracy of the data. However, ACLED acknowledges that these partners, along with witnesses and reporters, may not always be able to obtain comprehensive accounts of events or may be influenced by their own biases. This is particularly relevant in contexts where misinformation and disinformation are prevalent. To address this, ACLED adheres to a rigorous inclusion standard. While some aspects of reports on these events may remain speculative until triangulation is possible, ACLED’s EPO researchers actively seek out more complete and reliable information to ensure that the most reliable and up-to-date information is incorporated into the data. It is important to recognize that trusted sources, including local partners and reporters, need time to conduct thorough research and provide reliable information.

Using Trusted Sources

In the present Ethiopian context, there are a number of well-regarded and reliable sources of information. These include the independent Ethiopian Human Rights Commission; reporters at national newspapers; the social media accounts of humanitarians and practitioners; regional news agencies (reported in regional languages); and select radio, television, and online communities, which provide a wide range of information on regional and local conflict patterns. 

At the outset of the EPO project in early 2021, nearly half of all events come from national media sources. These include Addis Standard, Wazema Radio, and the government-affiliated Ethiopian Broadcasting Network, which provide detailed information on actors and fatality counts from local officials. These sources are coded in the data with a Source scale of ‘National.’ 

 The EPO also uses data from a network of development organizations based in-country that share information from local staff members’ anonymous reports of instances of violence. Such sources are useful in providing information in locations where media access has been restricted, and they appear in the data with ‘Undisclosed source’ listed in the Source column and a Source scale of ‘Local Partner-Other.’ 

A minority of coverage comes from international sources (e.g., Reuters or Sky News), including those with Ethiopian-language programming (e.g., BBC Amharic, VOA Amharic, or DW Amharic). Such sources provide information on conflicts and protests throughout the country and appear in the data with a Source scale of ‘International.’ 

The figure below depicts this source scale disaggregation in the data, illustrating the sourcing profile ACLED and EPO use across the country for coverage. While these sources are the primary source of information for any event coded by ACLED, multiple sources are reviewed and cross-checked in each instance to combat any misreporting and false information. When multiple sources are used to triangulate a single event, it is noted in the ‘Source’ column so that the triangulation of information is transparent. 

Regional-level Coverage

The EPO has also aimed to expand coverage to as many regionally specific sources as possible. Conflicts across Ethiopia have become highly regionalized, and many national sources of information are either limited in their abilities to cover the violence due to government restrictions on access to unstable areas or direct limitations generated by the insecurity. In turn, and after extensive in-country investigations of sources and sourcing, ACLED and the EPO integrated government communication bureau webpages (e.g. the Oromia region communication bureau webpage), the pages of political parties, and media in Tigrigna and Somali languages. This is in addition to a new stream of information fed through a network of development partners, as outlined above, reporting on instances of violence. The new EPO initiative brought about these changes, with a shift from the reliance on diaspora-based media to a prioritization of media and local sources who are based in-country in order to increase accuracy in identifying actors, locations, and fatality counts.

Coding Events and Reported Fatalities in a Challenging Information Environment 

EPO researchers review sources in English, Amharic, Somali, Tigrigna, and Afaan Oromo languages each week to code political violence and demonstration events in Ethiopia, in addition to information gathered from local partners. The methodology that ACLED and the EPO use to collect information has remained stable and is consistent with ACLED’s approach toward data collection worldwide.

Tracking fatalities is one of the most difficult aspects of conflict data collection in general, as fatality counts are frequently the most biased, inconsistent, and poorly documented components of conflict reporting, and this is especially true of active conflict environments impacted by high levels of mis/disinformation and severe access constraints. ACLED and EPO, therefore, take a conservative approach to fatality recording. 

If an event source does not reference fatalities, none are recorded. If a source mentions an unspecified number of fatalities, researchers consider where the event occurred and who was involved, and an estimate of three or 10 fatalities is coded depending on the circumstances of the event. For example: In an intense war zone where the event is more likely to have a high death toll, 10 fatalities are coded. In a low-conflict zone where the circumstances point to a death toll likely less than 10, a more conservative estimate of three fatalities is coded. If a source mentions a range of possible fatalities, the lowest number is recorded unless more information is available to substantiate a higher estimate. If a source includes a broad estimate like “hundreds of deaths,” and no additional information is included, 100 deaths are recorded for the event. 

Every ACLED event is associated with an acknowledged source. These sources are typically known, public, and open, with the addition — where necessary — of local, private sources. ACLED depends on these sources to produce an accurate description of political disorder events, and gauges the bias and robustness of sources through a rigorous assessment before inclusion. While conflict zones can produce highly biased sources, relevant information can be selectively extracted from such sources and gathered to reduce the impact of source biases and retain event information (for a recent discussion of bias in conflict data, see An Agenda for Addressing Bias in Conflict Data).

Often, standard rules around controlling the most biased parts of a report can lead to underreported fatalities, as fatalities are the most biased feature of event data (for more information, see FAQs: ACLED Fatality Methodology). They are subject to biases of exaggeration for media interest, reporter estimation bias, and source bias (i.e. which side of the conflict the information came from). For example: the Taliban had an active reporting service during its period as insurgents in Afghanistan (for more details, see The World According to the Taliban: New Data on Afghanistan). This service often reported information similar to other Afghan sources for decisive events and battles with government forces and also reported on the specific nature of small skirmishes. Yet, this source tended to report distorted fatality numbers that were largely sustained by government forces. As such, the fatality information was biased, but the event information was reliable.

Data additions

ACLED collects data in real time, but it is also a ‘living dataset,’ meaning that data are subject to updates, corrections, and additions on a weekly basis as new or better information emerges. Regular reviews are undertaken as more information is published and facts are established about historical conflict events.

 

For more information about ACLED’s core methodology, check the ACLED Resource Library.