The Armed Conflict Location Event Dataset (ACLED) is part of a general trend in social science to use the current benefits of global online media and information to disaggregate and track social phenomena. Through monitoring reports from multiple sources in multiple countries, datasets such as ACLED can move beyond the highly aggregated binary approach of classifying political violence seen in earlier studies and breakdown conflicts into a series of spatially and temporally discrete events (Gleditsch et al., 2013).
However, the adoption of media monitoring brings new challenges. Datasets relying on external sources are subject to the biases of their sources. Sources can introduce bias through both selective reporting and omission. Previous studies of media monitoring have found that selectivity often results in a geographical bias, with publications focussing on events near their base of operations and editorial offices (Barranco and Wisler, 1999). Similarly, in global reporting, reporting may be limited to where there are wire offices and the infrastructure to send information effectively (Woolley, 2000). Both of these dynamics have the potential to introduce a strong ‘urban bias’ into conflict reporting.
Kalyvas (2004) argues that urban bias is inherent in conflict reporting because security concerns compel observers to cluster within the confines of the main cities. The difficulty of gaining access to non-urban conflict zones is compounded by the fact that access to these areas is often contingent on perceived loyalty to the incumbent or insurgent forces who control the territory.
A review of ACLED events was conducted to see if there is evidence of urban bias in the data, and whether urban bias affects certain scales and types of publication. This was done by comparing the data in the location and municipal location columns of each event to a Geonames dataset of all urban areas with a population of 100,000 or over and coding matches with a 1 and non-matches with a 0. All events were filtered to have a Geoprecision code of 1, ensuring that only events with exact locations are included in the analysis.
Figure 1 compares the percentage of urban and non-urban events reported by different types of publication. An unexpected finding is that NGO reports do not have a significantly higher proportion of non-urban events (59.02%) than media sources (59.01%), given that NGO reports are often not subject to the limitations identified above. For example, NGO reports on political violence often take the form of interviews and in-depth investigations that take place after the event itself, meaning that NGO reporters should have easier access to the location of the conflict than journalists, who are compelled to report on the conflict as it is ongoing.
Reports conducted by the national government or international governmental organisations, such as the United Nations and African Union, have the highest percentage of non-urban events coded. This is likely due to the extra resources these organisations have at their disposal, such as armed escorts and guarded outposts in the middle of more rural conflict zones such as Eastern DRC and Darfur.
Figure 2 shows how different scales of publications compare on the proportion of urban and non-urban events reported. ‘International’ refers to publications from outside of Africa, ‘Regional’ refers to pan-African publications, ‘national’ refers to domestic publications and ‘local’ refers to publications with an editorial focus towards a certain region within a country.
International, regional, national and local sources show little variation in the proportion of urban and non-urban events coded. Non-urban events make up 55.86% and 57.90% of the events coded by international and national sources respectively, while 62.02% of events coded by local sources take place in non-urban locations. This lack of variance could possibly reflect an interdependence between the different scales of publication, with international media relying on national publications for information and national sources procuring their information from local sources.
Multiple sourced events (those coded using the information of multiple sources across a range of scales) show the highest percentage of non-urban events (73.23%). Studies evaluating the accuracy of data gathered from publication monitoring advocate the ‘triangulation’ of data using multiple sources to increase the accuracy and granularity of the data (Earl et al., 2004; Weidmann, 2013). Accessing multiple accounts of the same event will likely lead to better information concerning location. Indeed it may be a case of reverse causation where multiple sources are not needed to overcome urban bias but are in fact required to locate non-urban locations.
Barranco, José and Wisler, Dominique. 1999. Validity and Systematicity of Newspaper Data in Event Analysis. European Sociological Review, 15(3): 301-322.
Earl, Jennifer, Martin, Andrew, McCarthy, John D and Sarah A Soule. 2004. The Use of Newspaper Data in the Study of Collective Action. Annual Review of Sociology 30: 65-80.
Gleditsch, Kristian Skrede, Metternich, Nils W and Andrea Ruggeri. 2013. Data and progress in peace and conflict research. Journal of Peace Research 51: 301-314.
Kalyvas, Stathis N. 2004. The Urban Bias in Research on Civil Wars. Security Studies 13(3): 160-190.
Weidmann, Nils B. 2013. The higher the better? The limits of analytical resolution in conflict event datasets. Cooperation and Conflict 48(4): 567-576.
Woolley, John T. 2000. Using Media-Based Data in Studies of Politics. American Journal of Political Science, 44(1): 156-173.