Afghanistan

Coding decisions on the Taliban, external forces, the fall of Kabul, and more

Published on: 2 March 2023 | Last updated: 4 November 2024

The main conflict in Afghanistan began in 2001, following the US-led NATO invasion in the wake of the 11 September attacks and the subsequent overthrow of the Taliban government. In 2003, the Taliban announced it had officially regrouped. One year later, it declared that it had begun an insurgency under Mohammed Omar against both foreign forces and the US-installed Transitional Islamic State of Afghanistan, and later, the Islamic Republic of Afghanistan. The Taliban’s goal was to regain political power over the country and push out foreign forces. After nearly two decades of conflict, protracted negotiations between the Taliban and the US began in 2018. Notably, the negotiations excluded the US-backed Afghan government (New York Times, 29 February 2020). In February 2020, a peace deal was signed between the US and the Taliban, known as the Doha Agreement.¹ The agreement stipulated that both parties would refrain from using force against each other. The agreement also envisioned intra-Afghan dialogue and negotiations to start soon after. Despite the agreement and a permanent ceasefire between the Taliban and Afghan forces being one of the main agendas of the intra-Afghan peace talks, the level of fighting between Afghan state forces and the Taliban did not decline significantly.

The Taliban’s numbers are often bolstered by foreign fighters mainly from Pakistan, Uzbekistan, and Tajikistan, though the majority of them still identify as Pashtun (Foreign Policy, 15 June 2016). Taliban allies include both local and international groups such as the Haqqani Network and Al Qaeda, in addition to a number of other smaller militant groups (Combating Terrorism Center, September 2021). The Taliban allegedly receives financial support from a number of countries, and also relies heavily on profits from the opium trade (BBC, 28 August 2021). Airstrikes by both government and international forces have often targeted drug facilities due to these links.

In addition to the Taliban and its allies, the Islamic State (IS) is also active in Afghanistan. In 2015, IS announced the formation of an Afghanistan/South Asia affiliate which refers to itself as Islamic State of Iraq and the Levant – Khorasan Province (ISIL-KP). Its membership is composed of a number of foreign fighters as well as militants formerly affiliated with the Afghan Taliban, the Tehreek-i Taliban Pakistan (TTP), and other militant groups. The group pledged allegiance to former IS leader Abu Bakr Al Baghdadi, and its goal is to incorporate “Greater Khorasan” (a historical region encompassing parts of Central Asia, Afghanistan, and Pakistan) as a province of the imagined Islamic Caliphate declared by IS. For this reason, ACLED refers to the group as “Islamic State Khorasan Province (ISKP)” in accordance with the standard official name the group is referred to as. The presence of IS added a new element to the conflict in Afghanistan, as a common enemy for the Taliban, Afghan security forces, and NATO to combat. While the Taliban’s goal has often led to attempts to gain civilians’ trust, IS carries out suicide attacks in crowded areas, typically resulting in high civilian casualties (for more, see this ACLED report).

Since 2018, the Taliban has significantly expanded its control over the country, taking over rural areas and capturing provincial capitals (Long War Journal, 14 May 2021). In May 2021, the Taliban began to rapidly gain ground as US and NATO forces started the process of withdrawing from Afghanistan, as per the Doha Agreement. These efforts accelerated in early August, when the Taliban took control of several major cities in rapid succession (Guardian, 14 August 2021; Washington Post, 16 August 2021). On 15 August, the Taliban seized all administrative and military compounds in the capital city of Kabul, thereby taking over governance of Afghanistan (Associated Press, 16 August 2021). Earlier the same day, then-President Ashraf Ghani fled the country with several high-ranking government officials (Al Jazeera, 16 August 2021).

In late August, an anti-Taliban resistance began to form in the Panjshir valley in northeastern Afghanistan. The valley had previously housed the headquarters of anti-Taliban Northern Alliance forces, and is considered to be the historical heartland of anti-Taliban resistance. The current resistance — which has been referred to as the Second Resistance, or National Resistance Front (NRF) — is led by the son of the Northern Alliance’s leader, Ahmad Massoud. The NRF is composed of former Afghan security forces and local militias. On 6 September, the Taliban claimed that it had captured Panjshir province. However, through the end of 2021 and early 2022, clashes between the NRF and Taliban forces have continued in the province (Aamaj News, 25 December 2021; Business Standard, 8 January 2022).

On 7 September 2021, the Taliban formally announced a caretaker government led by one of its founding members, Mullah Muhammad Hassan (New York Times, 8 September 2021). Hassan assumed the role of prime minister, tasked with overseeing the day-to-day governance of the country, while Mawlawi Haibatullah Akhundzada was declared the supreme leader. Akhundzada was already appointed as the leader of the Taliban in 2016, and now as the supreme leader of the de facto Afghan government, he will likely have the final word on all political, religious, and military affairs (Gandhara, 7 September 2021).

Afghanistan presents unique methodological challenges for recording political violence. These primarily concern the extent of and access to media coverage, which has long suffered from reporting biases and limited access to Afghanistan’s most remote regions, as well as regions controlled by insurgent groups. The reporting biases often stem from journalists avoiding reporting on sensitive issues, such as violence against civilians and human rights violations, owing to intimidation and attacks from militants and Afghan state forces (Human Rights Watch, 10 April 2019). The situation is aggravated in remote provinces and regions controlled by insurgent groups as these areas lack proper security and access by foreign media (Human Rights Watch, 21 January 2015). This means reporting is only accessible via conflict actors themselves, which comes with its own reliability issues (explored in further detail in the sourcing section). Following the Taliban’s capture of Kabul, many sources have been stifled by the Taliban (DW, 27 August 2021). Additionally, reports of conflict events are often vague, using general terms such as “security forces” and nondescript references to “militants” or “terrorists,” in addition to aggregating what may be several distinct events into single provincial or district overviews. Furthermore, sources reporting on behalf of conflict parties, such as the Taliban’s Voice of Jihad (VOJ), may have incentives to share biased information (for more information, see Sourcing prior to the Doha Agreement and the implementation of RiV below). This methodology primer aims to outline ACLED’s efforts to address these challenges and to accurately capture manifestations of political violence in Afghanistan.

What does ACLED cover in Afghanistan?

ACLED collects data on political violence and demonstrations in Afghanistan, per ACLED’s core methodology (as outlined in the ACLED Codebook). Coverage of Afghanistan currently spans from January 2017 to the present, with continued weekly data releases. Efforts are always underway to expand and improve coverage through additional sourcing.

Which armed actors are recorded?

Marked changes took place in August 2021, when the Taliban effectively seized control of the country. The rebel Taliban force became the de facto government by overthrowing the government led by Ashraf Ghani, who first took office on 29 September 2014. This section describes actors coded in the country before and after these events.

Government, government-aligned, and Taliban actors before the fall of Kabul (pre-15 August 2021)

Government Forces: State forces in Afghanistan were composed of a number of state-sponsored security organizations. These were the military and police forces, as well as a third group known as the Afghan Local Police (ALP), which was a government-supplied paramilitary local police force supporting the Afghan National Police. Where sources only say “security forces” without specifying police or military forces, Military Forces of Afghanistan (2014-2021) has been coded by default. When possible, military and police forces were further broken down into prominent sub-actors for more advanced actor analysis. For police, these include wings such as:

Police Forces of Afghanistan (2014-2021) National Directorate of Security: the primary intelligence agency in Afghanistan. It was often the target of bombings. NDS is a common acronym for this actor.
Police Forces of Afghanistan (2014-2021) Sangorians: special forces designated to infiltrate Taliban cells with support from the NDS.
Police Forces of Afghanistan (2014-2021) Special Forces: police commandos/SWAT.

For military forces, this includes:

Military Forces of Afghanistan (2014-2021) Special Forces: often described as “commandos.”

When government operations were provided with air support, Military Forces of Afghanistan (2014-2021) was coded to account for the air force. There is no sub-actor for the air force; analysis of military air raids can be done by sorting the data on airstrikes made by Military Forces of Afghanistan (2014-2021). NATO was the only other actor conducting air operations in the country, and they are coded specifically when known to have been involved in an event.

Militia (Pro-Government): This actor — often described as “arbakis,” a Pashto word that loosely translates to “militia” — is used to describe militia groups that were unofficially backed by the Afghan government and/or NATO (Human Rights Watch, 12 September 2011). While some sources refer to the ALP as “arbaki,” others have chosen to treat the two groups as separate from one another, with the “arbakis” understood as an unofficial, yet distinct, version of the ALP (untrained, without official sanction). ACLED takes the latter approach.

Militant/Insurgent Groups: The most active insurgency in the country was led by the Taliban. ACLED also records a sub-actor for this group called Taliban – Red Unit, which is a special forces/commando unit. Despite the group’s prevalence, ACLED does not make assumptions that unidentified militant groups mentioned in sources are the Taliban, unless the event concerned occurred in an area where known Taliban operations were taking place at the time (for example, during the siege of Ghazni city). This is because other insurgent groups were also active in Afghanistan during this period of coverage. Other insurgent groups operating within Afghanistan are discussed in more detail below.

For Afghan Ministry of Defense (MOD) press releases, the actor Taliban and/or Islamic State Khorasan Province (ISKP) was often used for general “military operations” where the specific group was not mentioned. While the specific actor may be unknown, it is known that these large-scale operations were targeting one of these two groups. Unidentified Armed Group (Afghanistan) was used for all other cases where the armed group was not specifically named, since there were other groups operating within the country on a smaller scale.

External Forces: There were a number of other state military forces operating within Afghanistan, all of which can be sorted in the data by their interaction code ‘External/Other Forces’ (8) — a cross- ACLED code used for “other/external” forces, including state forces operating outside their own countries. The most prominent foreign force in Afghanistan was NATO: North Atlantic Treaty Organization, which is a joint force made up of several individual international militaries. It was often described by name, or simply as “foreign” forces in source information. Occasionally, sources would refer to a specific NATO member (for example, “supported by US airstrikes”). In such cases, the specific country’s military force was represented as an associated actor to NATO (for example, Military Forces of United States (2021-)). If NATO was supporting Afghan military forces, either NATO, or NATO and the specific country, were represented in the appropriate associated actor column. This format was also used for other coalition forces operating in other countries, such as the Global Coalition Against Daesh or Operation Restoring Hope (for more information, see this ACLED report on NATO in Afghanistan).

Government, government-aligned, and Taliban actors after the fall of Kabul (post-15 August 2021)

Since the fall of Kabul on 15 August 2021, the central Afghan government and its security forces have ceased to operate and control the country in a meaningful way. As the Taliban have been in de facto control of the country since 16 August 2021, they are now coded as the Government of Afghanistan. This determination does not denote legitimacy or international legal recognition, but rather acknowledges the fact that a distinct governing authority exists and exercises de facto control over significant portions of territory in a country.

While many members of the Afghan security forces have dispersed or outright joined the Taliban, some former government officials and security personnel have banded together in apparent opposition to Taliban rule. The full scope, coherence, and goals of this oppositional force, including the NRF, are still unknown as of early 2022. On one hand, these forces continue to form a coalition to fight against the Taliban. In addition to Panjshir, the NRF claims to have support bases in at least five other provinces (Diplomat, 15 December 2021). On the other hand, it is equally possible for these forces to strike a deal with the Taliban and to join a Taliban-controlled government. The Taliban administration has reached out to leaders of the NRF for talks offering amnesty for those who want to return to Afghanistan (Associated Press, 10 January 2022). Under these circumstances, ACLED chooses to code actors with caution in light of the volatile circumstances, and to refrain from assuming that coalitions have formed where the situation remains too fluid to make a lasting determination. As such, actors in Afghanistan after 15 August 2021 are coded as follows:

Taliban forces are no longer coded as “Taliban” with an inter code of ‘Rebel Groups’ (2) designating a rebel force. Instead, they are coded as Government of Afghanistan (2021-), Military Forces of Afghanistan (2021-), or Police Forces of Afghanistan (2021-), with an inter code of ‘State Forces’ (1) designating a (de facto) government actor. In addition, as and when possible, the Taliban military and police forces are further broken down into prominent sub-actors for more advanced actor analysis. The Taliban are coded as Military Forces of Afghanistan (2021-) Special Forces, when sources describe the Taliban forces involved in an incident as “special forces,” “Red Unit,” “Blood Unit,” and/or “Badri 313.”

A new actor, “Anti-Taliban Forces” with an inter code of ‘Rebel Groups’ (2) indicating a group trying to overthrow the de facto state, has been introduced. This is a generic actor, coded when unspecified resistance groups, explicitly reported to be formed to fight the Taliban, engage in violence and/or declare their support for larger, organized armed groups such as the National Resistance Front.
Former members of security forces banding together in new armed groups are coded as the associated actor Former Military Forces of Afghanistan (2021-) or Former Police Forces of Afghanistan (2021-) to the primary armed group in which they operate. This can be a named armed group, like the National Resistance Front, or an unspecified group, such as Anti-Taliban Forces or Unidentified Armed Group (Afghanistan).
Unless they take up arms, former security forces and government officials are coded with Civilians (Afghanistan) as the primary actor, and Former Government of Afghanistan (2021-) as the associated actor.

Unidentified Armed Group (Afghanistan) is used for all other cases where the armed group is not specifically named and is not covered by other generic armed actors (for example, when unspecified armed assailants target civilians).

As a result of the Taliban being coded as the de facto government force in Afghanistan, the coding of some sub-event types changes as well. Up to 15 August 2021, Taliban forces seizing control of a territory in a battle is coded as “Non-state actor overtakes territory.” After 15 August 2021, Taliban forces taking control of a territory in a battle is coded as “Government regains territory.”

Other significant actors, active both before and after 15 August 2021

The Khorasan branch of IS — coded as Islamic State Khorasan Province (ISKP) — was often the target of both Afghan security and NATO operations. While IS presence has been widespread across the country, the majority of events involving the group occur in the provinces of Nangarhar, followed by Kunar, Kabul, and Jowzjan. Because this group had goals that were similar to the Taliban’s, the two actors often battled for supremacy in areas where their control overlapped. Now that the Taliban have become the de facto government of Afghanistan, IS may challenge Taliban governance in areas where the group remains active.

Communal Militias are distinct from pro-government militias in Afghanistan. Communal militias do not work for or necessarily with state forces, but rather are armed communities/groups defending their larger communities and localities.

Smaller militant/insurgent groups operating throughout the country include:

Haqqani Network: a semi-autonomous entity that has operated both independently and in support of other Taliban groups, and has been active in the eastern provinces of the country. Some of the group’s leaders, including the chief Sirajuddin Haqqani, are high-level officials in the Taliban caretaker government. If the group becomes increasingly embedded within the Taliban, where distinguishing the two actors becomes impossible, this coding decision may be reviewed.
Al Qaeda: a global network of jihadists that formed as a response to the Soviet war in Afghanistan. US-led NATO forces invaded Afghanistan on the grounds that the Taliban were harboring Al Qaeda which was responsible for the 11 September attacks. Currently, they are active in provinces that border Pakistan, although their presence appears to be significantly lessened. They are allied with the Taliban.
Tehreek-i-Taliban Pakistan (TTP): a loosely tied umbrella network of militants based in the tribal regions of Pakistan. Divided into several factions, the network has close affiliations with Al Qaeda and is an ideological ally of the Afghan Taliban. Following the Taliban takeover of Kabul in August 2021, the TTP’s emir renewed his oath of allegiance to the Taliban.

To a lesser degree, small groups from Pakistan occasionally make an appearance within Afghanistan — possibly because they use the country as a base of operations. The most prominent of these groups is Lashkar-e-Islam (LeI), which often clashes with IS. Several political party militias have also been active in Afghanistan to a small extent, including Jamiat-e Islami and Hezbi Islami.

Pakistani state forces occasionally fire shells across the border (Durand Line) into Afghanistan, especially into the provinces of Kunar and Nangarhar, which border the former Federally Administered Tribal Areas (FATA). Shells are often fired by both Military Forces of Pakistan (2018-) and Police Forces of Pakistan (2018-) Frontier Corps. Although targets are rarely mentioned in sources for these events, and casualties are also rare in these events, it is assumed that the shelling acts as a deterrent for cross-border militant activity. In a small number of events, Pakistani and Afghan border forces have clashed.

How are events sourced?

Currently, each week ACLED researchers review approximately 110 sources in English, Dari/Farsi, Pashto, and Arabic in order to provide the most comprehensive database on political violence and demonstration activity in Afghanistan. Of those, nearly two-thirds of the sources are in English and the rest are in local languages (Dari/Farsi and Pashto), with some sources belonging to the Taliban and other insurgent groups sourced in Arabic.

Per ACLED methodology, the source column in the data will not note every single source/outlet that may have reported on a specific event. In the case of large-scale, high-profile events, the number of sources reporting on an event could be in the hundreds, with information often repeated across outlets. Rather, ACLED only codes the name of each source that contributed unique information used in the coding of the event. This means that if Source X reports on an event and the researcher uses information from this source to code the event, then Source X is coded as a source for the event. If the researcher comes across the same event reported by Source Y and Source Z, Source Y and Source Z will only be included in the source column if they contribute additional novel information that was not already captured by Source X (for more on ACLED methodology, see this sourcing primer or the ACLED Codebook).

ACLED relies on information from traditional media at the national, regional, and international levels. Approximately one-third of information used in ACLED’s coding of disorder in Afghanistan comes uniquely from traditional media sources. Each week, ACLED researchers review over 65 traditional media sources. Afghan national media makes up the larger portion of these traditional media sources. Often, these sources report on national news with more accurate event-based information. The country’s main national news outlets are covered, including, but not limited to, Bakhtar, Khaama Press, Tolo News, and Pajhwok. ACLED also relies on local or subnational media sources. Each week, ACLED researchers review over 10 subnational sources. Approximately 1% of information used in ACLED’s coding of disorder in Afghanistan comes uniquely from subnational media sources, such as local radio or TV stations. These sources typically cover reporting by local (district or provincial) authorities. Details around fatality numbers, victims of violent events, territorial transfers, etc. may differ from reporting by conflict actors. As such, when available, these sources are a helpful tool in corroborating information from other sources. Several international media sources are also reviewed, which tend to cover large-scale events and provide a broader picture of the conflict.

In addition, ACLED relies on local and global partners to supplement coverage. ACLED’s global partners include Front Line Defenders, which reports on attacks against human rights defenders, as well as the Aid Worker Security Database, which reports on attacks against aid workers.

Lastly, ACLED also sources information from ‘new media’ including LiveUAMap (described in further detail below, under Sourcing after the Doha Agreement before the fall of Kabul).

The media environment in Afghanistan has changed dramatically over the years. Recent changes began with the implementation of the Reduction in Violence (RiV) period, observed by Afghan, international, and Taliban forces, which culminated in the signing of the Doha Agreement in February 2020. The situation changed again following the Taliban takeover of Afghanistan in August 2021. The following section describes these changes and the strategies implemented to address the expected and realized sourcing gaps. ACLED sourcing is unique by context, and is regularly reviewed in order to adapt to evolving situations on the ground.

Sourcing prior to the Doha Agreement and the implementation of RiV

A key sourcing challenge in Afghanistan has been that, in addition to being the site of an ongoing and deadly conflict, much of this conflict has been concentrated in hard-to-access areas. The security situation has restricted media access to much of the country, particularly Afghanistan’s rural countryside. As of 2020, more than 70% of the population lived in rural areas, according to the World Bank (World Bank, 2021). In many cases, the primary conflict parties — the Afghan government and the Taliban — were the only sources of information for a particular event, reporting on this via the MOD and VOJ websites, respectively.

As parties to the conflict themselves, these groups may have had incentives to share biased information. As such, including these sources may introduce bias into the data, like inflated or deflated fatality counts. However, not including these sources likewise introduces another type of bias into the data, such as a distortion in where violence is happening, if events only appear in areas where independent sources have immediate access. Small-scale skirmishes or assaults in remote areas often occur in locations where other sources lack access, so limiting data collection only to information reported on by ‘independent sources’ would mean such events would be missed. This would lead such spaces to appear ‘peaceful’ when they may in fact be the site of conflict events.² As such, choosing to include or exclude information from these sources comes at a cost regardless of the final decision made.

ACLED finds that it is important to consider these events, and has determined that not all aspects of the information these sources reported was necessarily biased (more on this below). Afghan national media has also relied heavily on information from these sources, as they were not able to access all parts of the country. Nearly half of the information used in ACLED’s coding of disorder in Afghanistan prior to August 2021 is sourced from the MOD and VOJ. These sources are coded as source scale “Other” in the data.

ACLED found that these sources were more reliable in their coverage of core information, such as whether an event occurred or not. Even in cases where there may be a discrepancy between the two conflict parties in their reporting — e.g. the MOD denying the Taliban’s capturing of a district — other sources, such as local media, may corroborate the occurrence of the event. This is why, unless otherwise determined, events reported by these armed groups in the Afghan context were considered to have happened, although specific details were assumed to be less reliable.

There was greater discrepancy in this reporting around who the instigator of the violence was and regarding fatality estimates. However, ACLED does not systematically code the instigator of a clash, and our fatality estimates for Afghanistan are specific to the reports. Across all contexts, fatality numbers are the most biased aspect of reporting and ought to be considered carefully by users. In the Afghan context, where reports from the conflict parties themselves were heavily relied upon, this bias was even further exacerbated. To account for this, ACLED prioritized fatality estimates reported by sources other than the MOD or VOJ, whenever available, in addition to taking other precautionary measures (see the How does ACLED record fatalities in Afghanistan section below).

Additionally, to further reduce bias, ACLED relied on the New York Times’ weekly Afghan War Casualty Report, which contained information that the outlet’s local correspondents confirmed (and was much more akin to local reporting). This source was helpful in corroborating a number of VOJ reports, and in this way, also helped to ground ACLED’s inclusion of VOJ as a source.

While national media might use information from conflict parties in their own reporting at times, the biases of traditional media are still evident — namely, the appetite of traditional media consumers for larger-scale events, with smaller skirmishes still tending to go unreported/underreported by such media outlets. This meant that relying on national media alone would still result in missing a certain subset of events, even if national media also relied on reporting from the MOD and VOJ.

In a similar fashion, the independent website iCasualties was used to verify the deaths of NATO forces in Afghanistan when reported by other sources, such as VOJ. iCasualties sourced its information from news reports and press releases from the US Department of Defense, CENTCOM, the MNF, and the British Ministry of Defence.

Sourcing after the Doha Agreement before the fall of Kabul

As a result of the peace process beginning in early 2020 between the Taliban and the US, a shift in the MOD and VOJ’s reporting of violence was noticed — the sources stopped reporting on certain actions which they had previously reliably covered. This strategic change in reporting by the conflict parties resulted in the illusion of a significant decline in violence. As such, ACLED shifted its sourcing strategy from 2020 to mitigate this reporting bias. Data from a local partner, Afghan Peace Watch (APW), as well as reports from the media outlet Afghan Islamic Press, were added from 2020 onward.

The APW documented war-related incidents via a vast network of reporters across all Afghan provinces, with broad reach throughout the country at the district level. They began their coverage in early 2020, in line with the Taliban’s initial agreements to RiV amid the US peace talks. As a local conflict observatory, they were able to offer the kind of wide coverage that was previously only accessible via MOD/VOJ reporting. That the APW’s coverage began when the declining shift in MOD/VOJ coverage also started makes the partner a good source to counterbalance pre-2020 MOD/VOJ coverage.

Following the introduction of the APW in 2020, local partner coverage increased significantly from less than 1% for the period 2017 to 2019, to over 20% of all data for the year 2020 — and to nearly 7% for all data for the time period 2017 to 15 August 2021.

In addition, ‘new media’ sources were also increasingly covered. While ‘new media’ can be a powerful supplemental source in some contexts, its helpfulness in capturing trends across contexts varies widely. In Afghanistan, information gathering from such sources increased from less than 2% for the period 2017 to 2019, to over 6% in 2020. Part of this increase comes from ACLED capturing traditional media and government sources, such as Shamshad and MOD, via their social media presence where they occasionally shared additional information not found on their websites. Events based on these reports are coded with the original source name (e.g. Shamshad or MOD) as well as the social media platform on which the information was accessed (e.g. Twitter) in the source column.

Another part of this rise can be attributed to increased ‘new media’ coverage through LiveUAMap. LiveUAMap is an online news aggregator whose analysts and editors fact-check before displaying relevant conflict and disaster coverage through an interactive online map tool. ACLED has a partnership with LiveUAMap, and its information is used to supplement coverage of disorder across a number of countries of ACLED coverage. While the media sources used by LiveUAMap in the Afghan context are nearly identical to ACLED’s source list, their coverage of multiple verified Twitter accounts has been useful for covering events not captured by traditional or conflict party media. Most notably, these events include smaller-scale attacks by IS, and fighting between IS and the Taliban. Events coded based on information from LiveUAMap are noted as such in the source column.

Sourcing following the fall of Kabul

With the fall of Kabul on 15 August 2021, and the Taliban takeover of much of the country, the conflict between the Taliban and Afghan state forces aided by international forces largely concluded. Nonetheless, political violence in the country continues; sourcing these events in Afghanistan has become increasingly difficult for a number of reasons.

Firstly, international media and NGOs evacuated much of their staff from Afghanistan by the end of August 2021. Many Afghan media organizations and NGOs which are perceived by the Taliban as having collaborated with the former government have also attempted, with limited success, to evacuate their staff. Meanwhile, many staff members remaining in the country have gone into hiding as media organizations and NGOs either shuttered operations entirely or pivoted their reporting to be more aligned with Taliban sensitivities. For example, some national media outlets have reduced their reporting on violence perpetrated by Taliban forces against civilians, focusing instead on religious messaging.

As a result of these developments, the coverage of incidents in the country has become sparse. Reports center on Kabul, where most of the remaining journalists reside, while information on other regions is limited.

In addition, reporting by the MOD stopped on 15 August 2021. VOJ, while still intermittently operational, has changed its reporting, publishing more sporadically. VOJ information is now focused more on governance- and public works-related issues, in efforts to portray the Taliban’s achievements and legitimize their government.

With the Taliban in power, reports of violence perpetrated by members of the Taliban government are less likely to be widely covered by traditional media due to the fear of reprisal (VOA, 12 October 2021). Likewise, clashes with and attacks by anti-Taliban forces such as the NRF and competing groups like IS are likely to be underreported by traditional media as well. The Taliban seeks to project an image of itself as a security provider, denying the presence of any organized opposition or resistance (Al Arabiya, 5 November 2021).

In order to adapt to the changing nature of the media landscape in Afghanistan and to mitigate this underreporting, ACLED shifted its sourcing strategy after the fall of Kabul in August 2021. In addition to the already established list of sources outlined above, during the immediate aftermath of the fall of Kabul, over 40 additional ‘new media’ sources were tested and reviewed weekly. These included the Twitter accounts of reputable or verified journalists, civil society members, dissidents, international humanitarian organizations, members of the Taliban, and anti-Taliban forces. These ‘new media’ sources were test-sourced for a time period of at least four weeks, and were used to track unique events not reported elsewhere during this period. Based on this test sourcing process of these 40 sources, and following discussions about the reliability of these sources with external experts, ACLED identified and added to its regular weekly coverage nearly 20 of these ‘new media’ sources that were considered reliable and that consistently produced unique events not reported elsewhere.

Since the fall of Kabul, these ‘new media’ sources have provided individual reports of actions by the Taliban, including the destruction of physical structures (for religious/ethnic reasons), violence against civilians, targeted violence against journalists, and small-scale protests. Journalists’ Twitter accounts also report on and have helped in verifying clashes between Taliban and anti-Taliban forces, although with less granularity.

Meanwhile, Twitter accounts of anti-Taliban resistance groups such as the NRF, while not completely reliable in all aspects of their reporting due to bias, have proven useful in providing core information about whether or not an event has taken place. This assessment stems from a broader consultation with experts on the region as well as ACLED’s local partners. It was noted that both regional and Afghan national media have also relied on incident reporting (i.e. whether or not an event took place) from these anti-Taliban sources due to access limitations in regions where insurgencies are ongoing. Where possible, ACLED still corroborates the occurrence of events with reports from other local media or partners such as the APW. Aside from this, more specific details reported by these sources are coded with more discretion. In particular, due to the incentive these sources have to skew fatality counts, ACLED treats fatality reports from these sources as less reliable and takes steps to code them conservatively (see the How does ACLED record fatalities in Afghanistan section below).

While the long-term importance of ‘new media’ sources and the role they will play in producing unique information remains to be seen, as of early 2022, ‘new media’ sources have been invaluable in supplementing the dataset with core information about the occurrence of additional events in Afghanistan that are rarely reported elsewhere. Information gathering from ‘new media’ sources has increased, contributing to over 45% of ACLED’s Afghan coverage since the fall of Kabul, up from less than 5%.

In addition, data from an international partner, ExTrac, are also integrated into ACLED’s Afghanistan dataset. ExTrac documents incidents by “a range of violent extremist movements” by monitoring and tracking these movements’ official and unofficial communications. ExTrac data for Afghanistan cover IS-related incidents, beginning from August 2019. The data are useful for corroborating events where IS claims responsibility, including attacks on the Taliban and minority communities.

Finally, owing to reporting and access restrictions, ACLED expects to increasingly rely on local language sources, as these may be the only sources that have access to different regions of the country. In view of this expected change, ACLED has identified additional sources in Dari/Farsi and Pashto, like Etilaat e Roz, a national daily that has been consistently covering incidents even after the Taliban takeover of Kabul. New sources will continue to be identified and systematically integrated into ACLED’s Afghan coverage (for more on ACLED’s strategy around adding new sources, see this primer).

Where does violence take place, and how are locations recorded in Afghanistan?

While political violence in Afghanistan has occurred nationwide, the majority of events occur in the crescent surrounding the central mountainous area due to the geography of the Hindu Kush Mountains and their extending ranges. Conflict events have also been heavily clustered in provinces through which the highway linking Jalalabad-Kabul-Kandahar-Lashkargah passes, in the east of the country.

The Taliban takeover of Kabul in August 2021 brought an end to the nearly two-decade-long conflict between the Taliban and Afghan government backed by international forces. In the immediate aftermath, political violence in Afghanistan was concentrated around Kabul, Baghlan, Panjshir, Parwan, and Nangarhar. In Kabul, violence against civilians and armed clashes were recorded amid the evacuation and withdrawal of foreign forces. In the neighboring three provinces of Baghlan, Parwan, and Panjshir, armed clashes between the Taliban and anti-Taliban forces also took place. In the eastern province of Nangarhar, where IS is based, US forces launched airstrikes following the IS attack in Kabul on 26 August 2021.

It remains to be seen how Taliban rule will affect national patterns of political violence in the long term. As of early 2022, in addition to Kabul where the current Taliban administration is centered, political violence is clustered around Nangarhar and other provinces bordering Pakistan. These are the areas where IS predominantly operates. In addition, clusters of political violence are recorded around larger cities, such as Kabul, Jalalabad, and Herat, where anti-Taliban demonstrations have occurred in the aftermath of the Taliban takeover. Events are also recorded in Panjshir and its surrounding provinces, like Baghlan, as the remaining anti-Taliban forces, including the NRF, continue to resist Taliban rule.

Using boundary data provided by Mapbox, ACLED records events in Afghanistan across nearly 400 administrative divisions, with over 1,800 distinct locations georeferenced across the country. These include cities, towns, villages, and other populated places, as well as natural locations like mountains and valleys. Depending on the precision of the sources and the size of the recorded location (town, district, or province), researchers will select the appropriate geoprecision code to reflect the precision of the geocoding (for more on geoprecision coding, see the ACLED Codebook). Due to the remoteness of many parts of Afghanistan, in addition to varying levels of governmental presence in rural areas, the majority of sources will only identify the district or province in which an event has occurred (which are coded at geoprecision 2 or 3, respectively).

A geoprecision code of 2 is also used when sources give the name of a village or general area, yet researchers are unable to find coordinates for the specific location. In these cases, researchers code at the district capital and mention the village/area name in the event notes, unless further identifying information is provided to use in coding at a more natural location. Various tools, including GeoNames, OpenStreetMap, and OCHA atlases, are used in an attempt to find these specific locations. However, a wide range of transliterations and the presence of colloquial names for locations make this a difficult process. With this in mind, appropriately using geoprecision codes can help to control for any urban bias in the data, especially during analysis on conflict in rural versus urban settings.

Additionally, multiple unknown locations within the same district or province are combined into a single event if all other details are the same. For example, “On 7 August 2018, 13 Taliban militants were killed by Afghan security forces in the Balakhel and Paeen areas of Kot district, Nangarhar.” would be coded as two events — one in Balakhel and one in Paeen. However, if for any reason the Balakhel and Paeen areas could not be specifically located (due to different spellings, local slang, name changes, etc.), a single event would be coded in Kot district (geoprecision 2), as opposed to coding two identical events, each aggregated to Kot district to denote these events.

Along these lines, in an effort to code conservatively, similar events involving the same actors in the same location (with the same geoprecision) on the same day are considered to be the same event, even if casualties differ, due to the wide discrepancy in reported fatalities (for further information, see How does ACLED record fatalities in Afghanistan? below). Events coded with geoprecision 1 or 2 take precedence over events with geoprecision 3 (where geographic information is unclear and only the province is known). For example, if one report states that “21 Taliban militants were killed in military operations in Ghazni province,” and another report says that “18 Taliban militants were killed by soldiers in Ab Band district of Ghazni province” on the same day, the latter will be assumed to be the same event and chosen. This is due to the latter report’s more specific location given that the former “21 militants killed” event could in fact be the “18 Taliban militants killed” event occurring in Ab Band specifically, with fatality numbers not matching. This is not unlikely given the nature of fatality estimates.

How does ACLED record fatalities in Afghanistan?

Fatality counting in the Afghan context faces several obstacles which make efforts to provide reliable estimates difficult. Scarce or biased reporting, as well as limited media access to the sites of violence, may result in substantially different fatality estimates arising from the same event, uncertain figures, or one-sided coverage of conflict events in certain areas.

Both the Taliban and MOD had reason to avoid reporting their own losses while also inflating the losses of the other side; the war they were fighting was protracted and required the continual support of the population and ongoing recruitment (New York Times, 21 September 2018). As such, fatality counts reported by independent or local sources took precedence when determining how many fatalities to code.

To avoid artificially increasing the number of reported fatalities, ACLED had taken steps to ensure that fatality estimates related to the conflict in Afghanistan are the most accurate possible and least subject to media biases. As nearly two-thirds of the information used in the Afghan dataset comes from the MOD and VOJ, and the fact that these sources were often the only ones reporting on events, recording fatalities as zero in all cases when the MOD and/or VOJ were the only source for an event would significantly undercount fatalities. So, in events where the MOD and VOJ were the only sources of information, ACLED deferred to coding estimates from the MOD rather than VOJ.

Of the two sources, ACLED assumed that the Afghan government had more oversight than the Taliban when it came to claims around fatalities, especially in larger attacks. Fatality estimates from VOJ are assumed to be too biased to include. This means that if VOJ is the only source of information for an event, fatalities for that event are recorded as zero.³ This protocol ensures that the total fatality estimates from these data are closer to reality, yet still remain conservative. While the MOD fatality estimates included may be inflated, this is offset by the undercounting of fatalities stemming from events in which VOJ is the only source of information. To reiterate, VOJ reports are coded with zero fatalities. This process is in line with ACLED methodology around fatalities (for more, see this methodology primer), which holds that it is best to work with fatality counts as aggregates rather than in understanding specific numbers of fatalities per event.

Following the capture of Kabul by the Taliban, the MOD ceased operations. While VOJ is still intermittently operational, ACLED assumes that fatality estimates from VOJ are still too biased to be included. Also, since one of the conflict parties, the former Afghan government, is no longer a prominent active actor, it can be assumed that VOJ will lack reporting on armed clashes. Any reporting on clashes and fatalities by VOJ can be assumed to be biased as the publication still acts as a mouthpiece for the Taliban and does not represent a formal governing body with oversight. If and when formal channels of government communication are established by the Taliban and begin reporting on events, ACLED will review the source and appropriate methodology will be established to account for the fatalities reported by those sources. As such, fatality counts reported by independent or local sources will continue to take precedence when determining how many fatalities to code.

Meanwhile, sources belonging to pro-NRF groups and other anti-Taliban resistance groups have emerged since the fall of Kabul. While being fairly reliable in their coverage of core information, such as the occurrence of an event, these sources have been found to be less reliable in their reporting of specific details. In particular, based on consultations with external experts, it was determined that these sources have an occasional tendency to either overcount or undercount fatalities. Therefore, ACLED has taken additional steps to code these fatality numbers conservatively, without completely disregarding them. When fatality reports from these sources are vague and mention that there were “deaths and injuries” or “losses,” or if they report more than three fatalities, researchers code an estimated fatality count of three. This adheres to ACLED’s broader methodology on coding unspecified fatality numbers (more below, also see FAQs: ACLED Fatality Methodology). If these sources report three or fewer fatalities, then the exact number is coded.

Data from ACLED’s partner, ExTrac, include fatality numbers claimed by IS. In the past, claims published by IS through several of its mouthpieces, including the news agency Amaq, have been “typically correct” (New York Times, 8 June 2017). Comparisons of IS fatality claims to other media reports have shown to be mostly in agreement on fatality estimates for the same events. IS-claimed fatality numbers are therefore coded per ACLED’s general fatality methodology, which is to code the most conservative reported number when different fatality counts are available (see FAQs: ACLED Fatality Methodology).

Meanwhile, across all sources, researchers triangulate reported fatality counts to always select the most conservative estimate available, unless more recent or verified information is released. For example, Khaama Press reports that “On 13 October 2018, 12 civilians were killed by a NATO airstrike in Ghazni city,” while Agence France Presse reports that 10 were killed in the same incident. Here, 10 would be chosen as the more conservative estimate between the two sources. However, if the UN sends a team to investigate and reports that 11 civilians were killed, the event will be updated to include this new reliable report.

Up till the fall of Kabul, if fatality estimates were unknown — and this happens often as many reports tend to be vague and only mention the occurrence of “deaths and injuries” or “losses” — ACLED used a standard estimate of 10 for Afghanistan, or three when the event was known or likely to have caused fewer than 10 fatalities (e.g. the bombing of a motorcycle resulting in some fatalities, or a small skirmish at a remote checkpoint).

Since the fall of Kabul, Afghanistan no longer meets ACLED’s criteria of being a warzone.⁴

Hence, when reports are vague and only mention the occurrence of “deaths and injuries” or “losses,” ACLED researchers now code a standard estimate of three for Afghanistan if no other guiding information is available. Additional intermediate estimates are used to capture other inexact reported figures, such as dozens, scores, etc.

Fatalities by Actor: ACLED does not code fatality figures according to which group suffered fatalities because many source reports do not offer this level of detail and, when they do, this information is often biased. Instead, ACLED codes the total number of deaths reported from an event in totality, which may at times include civilians killed as ‘collateral damage.’ This means that aggregate estimates of “civilian fatalities”⁵

stemming from events in which civilians were directly targeted do not include civilians that may have died as ‘collateral damage’ during fighting between armed groups or as a result of the remote targeting of armed groups (e.g. an airstrike hitting a militant position that also kills civilians). Any analysis of the total impact on civilians will hence require additional modification on the part of the user, along with a number of assumptions as to how to attempt to disaggregate these total event fatality numbers.

Splitting Fatalities: Many sources will release a single fatality total referring to events occurring across a number of locations; for example, “On 2 December, 33 Islamic State militants were killed during NATO airstrikes in the Rodat, Kot, and Khogyani districts of Nangarhar.” In this case, the fatalities are split equally over three events (in Rodat, Kot, and Khogyani) occuring on the same day, with 11 fatalities attributed to each location. Odd number totals will be split as evenly as possible; for example, 34 fatalities over three events would be split into 11, 11, and 12 (for more on this, see the ACLED Codebook). The notes will be the same for each event and will state that the fatalities have been split. It is important to remember that though the notes column entries may match across these events, these are not ‘duplicates’ but rather are unique events with different locations or dates coded.