14 August 2024: our work, coi, standards, acled, conflict, howtodescribeaconflict, methods-series

ACLED's Conflict Exposure – One Measure to Describe a Conflict

The introduction of this measure followed the addition of a new dataset variable called ‘civilian targeting’ in 2023. Both additions provide greater scope for analysing and interpreting ACLED data in terms of the impact of conflicts on civilians. As this aspect plays a central role in many COI products dealing with security-related topics, this methodological blog post is dedicated to the discussion of this new measure. This blog post is the first in a series of methodological blog posts that will generally address the question of How to best describe a conflict in the context of COI.

ACLED developed its conflict exposure measure in collaboration with WorldPop[1] to provide insight into how civilians are affected by conflict. Using fine-grained population data in combination with event-based conflict data, the measure offers estimates of the impact on populations based on their proximity to events, the type of event and the specific armed group. Thereby, the measure estimates the number of people living within 1, 2 and 5 kilometres of conflict events, aiming to provide a detailed view of the demographic and geographic exposure of civilians to conflict.

Population exposure statistics provide additional insights into traditional conflict measures such as the number of security-related incidents (indicating the nature of a conflict), the number of fatalities (indicating the intensity) or the number of affected areas (indicating the concentration of a conflict). The term ‘exposure’ refers to the impact and level of harm caused by conflicts, quantified by estimating the number of people affected in a given area over a given period of time. ACLED defines being exposed to conflict as living “in an area of active disorder or unrest” and specifies that individuals can be adversely affected by exposure to conflict in a number of ways: “they may be directly injured; they may find themselves in active conflict; they and their group may be targeted; or they may be affected by the destruction of their village, neighbourhood or town” (ACLED, n.d.).

ACLED's new measure allows exposure levels to be calculated by incident, location, time, actor and actor type, but also aggregated to monitor national and global developments over time. ACLED also provides a “best estimate” based on the default measures for the 1, 2 or 5 km radius of an event, depending on its type (5 km in case of battles, explosions/remote violence, and violence against civilians with at least one reported fatality; 2 km in case of violence against civilians with no reported fatality, and riots). The conflict exposure measure can be accessed via ACLED’s data files, where the different population columns with data on the different buffer sizes (1 km, 2 km, 5 km, best) can be added as additional variables, or online via the conflict exposure calculator available here.

A closer look at the methodology behind ACLED’s conflict exposure measure

Unrepresented migration and refugee flows

Population figures for areas affected by security-related events are approximations based on certain assumptions about the impact on nearby residents. The conflict exposure estimates depend on various factors, such as the location of the event and methods for estimating population. ACLED specifies, “these data are extrapolations from census, remote sensing, and other information” (ACLED, n.d.). Despite ongoing efforts, at the time being, the population data is only updated once a year and does therefore not reflect dynamic migration and refugee flows, or IDP movements, which are common in areas of conflict. (More detailed information on WorldPop’s methodology, can be found here.)

In brief, dynamic changes in population numbers are not immediately reflected in the population figures, and thus are not accounted for in the number of conflict exposed individuals.

Efforts to avoid double-counting

The ACLED conflict exposure measure aims to avoid over-counting the number of people affected. To avoid double counting, multiple security-related incidents documented in the same location within one single reporting period do not result in an increase of the exposure measure. It only increases when an additional time period or location is added. This means that if, for example, 10,000 people are exposed to a security-related incident documented for location X, and there are 5 such incidents at that location in the corresponding reporting period, the exposed population is still counted as 10,000 people, not 50,000 people. A person who is affected by several incidents in their neighbourhood within a reporting period is thus always only counted once for that period.

Moreover, the use of buffer zones also prevents people within the selected radius of multiple events from being counted more than once. This means that if an event is recorded in the immediate vicinity of the location where another event took place during the same time period, the buffer radii are modified to prevent people from being counted more than once[2]. In areas of high conflict density, the radii may therefore be smaller than specified, as the exposed population to be documented will be counted anyway due to another security-related incident nearby. (More detailed information on the modified buffer zones based on the so-called Voronoi tessellation technique can be found here.)

In brief, each person recorded as exposed is only counted once within a specified time period, regardless of the intensity or frequency of the exposure.

The geographical precision caveat

In the ACLED data, each incident is recorded with an exact geographic location. However, as the exact location is not known or reported for all incidents, the precision of geographic data in the ACLED data varies among incidents. In cases of incomplete geographic information, a town may be used to represent a region, or the provincial capital may be used when only the incident’s province is reported. ACLED records the precision level along with location data to reflect the fact that the precise location of the incident may not be known. For instance, a precision level of “3” means that the provincial capital is recorded as incident location when only the province is known. A precision level of “1” means that the exact village, town or city is reported and recorded. This method of handling imprecise data can impact the conflict exposure measure in several ways:

• Overestimation: Administrative centers like province or district capitals are usually more densely populated compared to remote locations. When such centers are recorded because only the province is known, the exposed population is calculated based on that center’s population. The actual incident location might have a smaller population.
• Underestimation: The conflict exposure measure counts each location’s population only once for any given time period. For instance, when examining a year, the population of an area is counted only once even when there are multiple incidents in the same area. The lack of precise reporting on incident locations leads to the same locations (e.g. region’s capital) being recorded instead of the actual diverse locations. Thus, instead of counting the different populations as exposed, the population of the capital is counted. When multiple events take place in various places in a province, but the exact locations are unknown, the capital is selected multiple times, and its population is counted only once. If the reporting had been more precise, the population of several places would have been counted. (Within the respective events’ radii.)

As mentioned above, the conflict exposure measure uses various radii to estimate the affected population (like 2 km for riots, 5 km for battles). However, for geographic precision levels of 2 or 3, the distance between an incident’s actual location and the recorded location is often far greater than 5 km. Even for the highest precision level of 1, when the city or village is known and recorded, it is usually the geodata of the city center that is used, rather than that of the actual location of the event. This means that even for data with high precision the recorded location can deviate by several kilometers from the actual location.

In addition to geographic precision levels, a second factor concerning geo-data is ACLED’s approach in handling the recording of locations, which can vary from one city to another:

“In selected large cities with activity dispersed over many neighborhoods, locations are further specified to predefined subsections within a city to prevent excessive aggregation of events to a single city location. In such cases, locations are recorded as: City Name - District name (e.g. Mosul - Old City)” (ACLED, 9 November 2023, p. 35 )

When this approach is used, such as in Aden, incidents are recorded in several subsections. As a result, the subsections’ centres and their surrounding populations are included in the conflict exposure measure. In cities where no subsections are used, like Kabul, the city centre is used for each incident in that city, and its population in the conflict exposure measure is included only once. Higher specificity in incident locations may thus lead to higher numbers in the conflict exposure measure.

Figure 1: Illustration of recorded locations in Aden and Kabul (map excerpt). Data source: ACLED Curated Data Files for Asia-Pacific and for Middle East (19 July 2024). Map source: © OpenStreetMap contributors.

The ACLED data contains records for numerous subsections in Aden (Yemen), like Aden - Al Maalla, Aden - Al Burayqah, Aden – At Tawahi etc, but a rather small number of distinct locations in the city of Kabul (Afghanistan). The vast majority of incidents in Kabul City are recorded at a single location in the centre of Kabul. The maps provided above may help to better illustrate how the recording of locations of security-related incidents may affect the population estimated to be exposed to them: The map on the left shows all distinct locations where ACLED recorded incidents coded as explosions/remote violence in Aden over the past 6 years; the map on the right shows all distinct locations where ACLED recorded incidents coded as explosions/remote violence in Kabul over the same time period. The data contains 157 such incidents at 55 distinct recorded locations in Aden, but 516 such incidents at only 23 locations in Kabul. 372 incidents have the recorded location “Kabul”. The greater diversity in the recording of the locations of the reported incidents in Aden means that more geographical areas are included in the exposure calculations, which may lead to higher estimates of exposed populations in a given time period.

In brief, the method of how incident locations are recorded with varying precision affects the conflict exposure measure.

Validity and usefulness of the measure in the context of COI

As with any (quantitative) data, knowing how it was collected and what it actually measures is key. In the case of conflict exposure, the fact that a person can only be counted once per reporting period, even if they are affected by several different conflict events within that period, is central to the interpretation of ACLED’s conflict exposure data. When comparing two countries in the same period Y, the same percentage of the population may be identified as exposed to conflict, even though the frequency of events in country A may be much higher than in country B, as the conflict exposure measure does not take into account the frequency of events; yet this is an aspect that strongly influences the situation and level of exposure of an individual. Likewise – unless otherwise specified by the selection of specific event types or actors – the conflict exposure measure does not give any indication of the nature of the conflict, meaning that the events recorded for country A may be mainly demonstrations, whereas in country B the affected population may be exposed to battles, explosions, etc. Additionally, methods and uncertainty levels when estimating population numbers differ from country to country, as does the reporting of incidents by ACLED’s sources. All this suggests that comparing different countries and different conflicts based solely on the percentage of the population exposed is not necessarily meaningful and, above all, can be misleading.

To illustrate this, if we use the ACLED conflict exposure calculator to determine the conflict exposure of the population of Pakistan and Austria in 2022 (without specifying any particular event type or actor), we obtain quite similar percentages[3], with Austria (35%) even showing a slightly higher percentage than Pakistan (33%) (see Figure 2).

Figure 2: Conflict exposure between 1 January – 31 December 2022 for Pakistan and Austria, with no specification regarding event types or actors. Source: ACLED, Conflict Exposure Calculator, as of 29 July 2024, https://acleddata.com/conflict-exposure/#calculator

However, a quite different picture emerges when we take a closer look at the nature of the described conflict by taking into consideration the various types of conflict events. As shown in Figure 3, the percentage of the conflict exposed populations in Austria in 2022 is down to 0%, when specifying the event types to battles, explosions/remote violence and violence against civilians:

Figure 3: Conflict exposure between 1 January – 31 December 2022 for Austria, conflict event type specified to battles, explosions/remote violence and violence against civilians. Source: ACLED, Conflict Exposure Calculator, as of 29 July 2024, https://acleddata.com/conflict-exposure/#calculator

In contrast to a comparison of various countries, comparing equivalent time periods for the same country, region or city, provides a much more meaningful insight illustrating a development over time (see Figure 4):

Figure 4: Conflict exposure between 1 January – 31 December 2022 and 1 January – 31 December 2023 for Pakistan, with no specification regarding event types or actors. Source: ACLED, Conflict Exposure Calculator, as of 29 July 2024, https://acleddata.com/conflict-exposure/#calculator

Conclusion

While incident and fatality figures can be used to describe the extent and intensity of conflicts, ACLED’s conflict exposure measure has the capacity to illustrate their impact on civilians, as indicated by the estimated number of individuals exposed. To be considered exposed to the conflict, civilians need not be the actual or primary target of a security incident, nor need they be injured or killed in such an incident, as there are other more indirect forms of exposure as well. Therefore, we consider the new measure of conflict exposure is a valuable addition to the ACLED figures on the number of incidents, the targeting of civilians, and general (not only civilian) fatalities. Particularly in the context of quantitative descriptions of the security situation, conflict exposure can provide a useful add-on to COI reports on security-related issues, as it allows a better understanding of the varying effects of different conflict types and actors on civilian populations.

In the context of COI research, we, however, encourage to keep in mind the limitations of the measure. These include, as mentioned above, the lack of accounting for dynamic population movements, the implications of efforts to avoid double-counting, and the risk of over- or underestimation due to the varying levels of geographic precision in conflict event data. When incorporating the conflict exposure measure into COI products, we propose that comparing equivalent time periods for the same location/area is the most suitable approach.

When describing a security situation or a conflict quantitatively, several measures need to be combined in order to obtain a more nuanced picture. Upcoming blog posts will examine other measures offered by quantitative conflict data collections. The next will focus on ACLED’s “civilian targeting” variable.

---

[1] WorldPop is a research group that provides subnational population data to all UN agencies.

[2] Note: It is not clear from the current description of the measure at what point these buffer adjustments take effect. When using the Conflict Exposure Calculator, the population figures for each location remain the same, whether a single location is queried for one day, or multiple locations and days with incidents in close proximity to each other. This would suggest that the buffer zones and population data for each location have been calculated once for a longer period of time (taking into account nearby incidents), rather than being specifically calculated for each query. For specific queries, the exposed population may therefore be underestimated.

[3] The numbers presented in Figure 2 suggest that ACLED’s calculator uses the 5 km radius as the basis for determining the percentage of the exposed population.