New and Emerging Forms of Violence Data for Crisis Response: A Comparative Analysis in Kenya

Lead Research Organisation: Institute of Development Studies
Department Name: Research Department


The project will produce the first robust evidence base on the opportunities and limitations of social media data on violence reporting to inform UK emergency and crisis response. These responses include targeted humanitarian support to vulnerable and conflict-affected populations, development of rapid conflict and risk assessments to inform policy and strategic action, and support to political and other reconciliation efforts in the medium-term (DfID, 2009, 2010). Effective UK Government crisis and emergency response increasingly depends on the availability of timely, reliable data on political violence, to determine the scale and dimensions of crises and tailor responses (UKAid, 2013). While social media reports of violence can inform the design, targeting, and geography of crisis response, there is limited robust research on their reliability and comprehensiveness.

Research highlights significant 'technology gaps' between members at margins of society, and those whose voices are amplified by social media (Perera, 2015); reflecting concerns that social media does not equally reflect experiences of the most vulnerable, including women and girls, or populations with limited digital access, who are often key crisis response targets. Conventional media also reproduce certain biases in selection and coverage (Wigmore-Shepherd, 2015). Nevertheless, little research has rigorously tested the nature or extent of data gaps, meaning policymakers develop responses without robust evidence on data reliability and comprehensiveness.

This project addresses this gap, by testing reliability and comprehensiveness of social media data, against conventional media reporting of violence in a real-time context: the August 2017 Kenyan elections. It will identify opportunities new data provide for policy, and what limitations restrict usability, along three dimensions: 1) reporting timeliness; 2) targeting of crisis response; 3) geographies of violence risk. Building on extensive social media use in Kenya, and a history of violence reporting via social media (Meier, 2008) the case facilitates a test of social media data in a promising context.

Two sources of political violence data will be comparatively mapped and analysed: 1) reports from the Uwiano social media and crowdsourced digital platform in Kenya; 2) reports coded from published media sources by the Armed Conflict Location & Event Dataset (ACLED). Data will be collected for six months (four months prior to, and two months following, elections). Findings will be discussed, validated and ground-truthed in workshops in Nairobi with policy, research, and civil society experts on Kenyan political violence, to assess comprehensiveness of both data sources and implications for response.

Research impact and uptake will be ensured through the central role of the Department for International Development (DfID) and Foreign and Commonwealth Office (FCO) staff, as non-academic members of the project team. This will ensure maximal policy relevance through data use and analysis in accordance with UK Government priorities; support research uptake and impact through targeted integration of evidence into specific UK Government analytic, policy and planning instruments; and facilitate performance assessments of complementary UK Government programming. The project will ensure impact and value for money through creation of a robust evidence base on use of new and emerging data; and systematic mapping of strengths, weaknesses and opportunities of discrete data in a real-world comparison.

To our knowledge, the proposed study represents the first attempt to rigorously and systematically map these opportunities and limitations in comparative context, and thus provide a robust evidence base for learning and deployment around digital violence monitoring platforms.

Planned Impact

In addition to academic beneficiaries discussed above, we expect the project to have important policy and social impacts. At policy level, we expect the research to identify entry points for policymakers to engage with new and emerging forms of data in emergency and crisis response, with greater confidence in the reliability and actionability of these sources. We also aim to contribute to longer-term UK, Kenyan and international policy agendas on violence monitoring, and electoral violence in particular, by producing research that robustly demonstrates the opportunities and limitations of new and emerging forms of data in this field.

We expect this research to directly inform policy decisions in several ways. First, we expect findings to inform the scale and type of support offered to social media monitoring platforms by UK Government agencies, including DfID country offices, global conflict profession, and innovation and evidence units. Evidence robustly mapping opportunities and limitations of new and emerging forms of data will inform the expectations and requirements of support to these initiatives. The project will offer insights into where new and emerging data can be used most effectively, and where it may be most robustly paired with other data collection methodologies. DfID Kenya already supports violence monitoring platforms which directly seek to identify, track and respond to violence in real-time. Related programmes include support to hate-speech monitoring initiatives which track incidents of hate speech and incitement to violence on social media. DfID and other donors' wider portfolio of programming in Kenya and beyond increasingly engages with new and emerging forms of data, but may lack the robust evidence needed to identify gaps and limitations, as well as opportunities and strengths, in making programmatic and strategic decisions.

Second, we expect the findings to contribute evidence on the soundness of different methodologies for monitoring and tracking violence among FCO analysts engaged in real-time conflict monitoring, and channelling assessments into policy. Government analysts rely on a wealth of data sources to track changing political circumstances, and assess the nature and scale of emergency and crisis needs and appropriate responses. This project will directly support those assessments by providing a robust evidence base for the reliability, actionability and usability of these sources of data in monitoring evolving, dynamic, real-world crises. In this way, the project findings will feed into policy decisions regarding the implications and consequences of the use of these data sources for emergency and crisis response. For example, the project findings will directly address the question of when, how and in what form can policymakers confidently rely on new and emerging forms of data, and how reliable and actionable are these sources for making decisions on the allocation of emergency and crisis response resources? The project will systematically and comprehensively outline the strengths, opportunities and weaknesses of new and emerging forms of data, equipping analysts and policymakers with a robust evidence base which maps both the extent and nature of gaps or limitations, as well as illustrating key strengths and opportunities.

At the social level, we aim to use the project's findings to support local civil society peacebuilding, digital, and social media initiatives to strengthen and improve associated platforms. Relevant programmes include the Nairobi-based iHub and Article 19, with which the project team have direct links. We will actively seek to engage with stakeholders at every stage of design and implementation, and share findings as widely as possible, in order to equip local initiatives and civil society organisations with the knowledge necessary to make strong, evidence-based advocacy claims using new and emerging forms of data in a reliable and robust way.
Description This research project has generated a number of new and significant findings, and developed a novel and original methodology to compare different media sources reporting on violent events, using the 2017 Kenyan elections as a case study. The study has compared reports generated through traditional media, using the ACLED database (a much used source for comparing traditional media reports), to Twitter reports of violence, using a novel method for combing Twitter for violence related tweets, using an algorithm developed by the Department of Informatics at the University of Sussex (named Method 52), along three dimensions: 1) Geography and geographical coverage 2) Temporality- timeliness, temporal coverage and time precision and 3) Targeting/representativeness.

The results provide new and valuable insights on the relative advantage of different reporting systems on political violence, their usefulness for violence monitoring and early response systems, as well as humanitarian interventions, as well as for policy-makers and academic research. The key findings of the study are:

1. Investing in traditional media remains crucial for violence monitoring. Our study has shown that traditional media retain strong value for reporting on violence, as they provide more temporally consistent, detailed, and quality reports on violent events than Twitter. However, the quality, consistency and timeliness of traditional media reporting on violence is dependent on the size, quality and diversity of the traditional media sector, which can vary significantly across and within countries, and over time. In an age of widespread de-legitimisation of traditional media, these results show that investing in a high quality, transparent and democratic media sector remains a central policy priority.

2. Investing in Twitter based monitoring of violent events is particularly relevant for protests and riots, and events that focus the attention of Twitter users. The study has shown that Twitter's coverage varies significantly over time, geography, and type of event, all resulting from the underlying behaviour of Twitter users. Twitter is particularly informative for reporting on protests and riots, and more generally for events that focus the attention of Twitter users. Further research is required to understand how the underlying characteristics and behaviour of Twitter users influence reporting patterns and biases.

3. Policy-makers and practitioners should carefully weigh the advantages and disadvantages of different monitoring systems before implementation, and favour combined approaches. Rather than being seen as alternative or competing sources, 'old' and 'new' media sources should be conceived as complementary as they both provide unique information that, together, paint a fuller picture of political violence. They are closely interdependent in the new media ecosystem, and both carry inherent biases and limitations, which should be weighted carefully before choosing a particular system. Particular attention should be paid to questions of the accessibility and representativeness of monitoring systems.

1. Policy-makers and practitioners should invest in the integration of monitoring systems into wider crisis response mechanisms. The proliferation of 'old' and 'new' media monitoring systems and violence reporting systems does not entail that these are well integrated in rapid-response systems. Policy-makers and practitioners should invest in reinforcing and centralising information distribution to rapid-response actors.
Exploitation Route First, the results of this study will be of high value for policy-makers and practitioners involved in violence monitoring, violence prevention and rapid-response to violence, as well as humanitarian actors operating in contexts marked by political violence. The study points out the relative strengths of different reporting systems, according to three key dimensions of 1) Temporality, 2) Geographical coverage and 3) Representativeness. The results allow such actors to tailor their monitoring and reporting strategies to the types of violent events that they face, and, when financial means allow it, invest in combined approaches.

Furthermore, the study is of value to academics and policy-makers in search of instruments to gather high quality data on violence, and can help them choose methods for research projects or violence monitoring systems. Given the sunk costs of investing in particular methodologies, the study results give clear guidance as to what type of data collection method is most appropriate for different types of violence - from riots and protests, to organised violence or strategic developments. The study also comprises a detailed review - in the first Working Paper - of the different biases that are present in different data collection systems.
Sectors Digital/Communication/Information Technologies (including Software)


Democracy and Justice

Security and Diplomacy

Description A presentation of the research rationale, methodology and preliminary findings was made to the ESRC on December 8, 2017, at the British Academy in London. The second project workshop, gathering leading actors in the field of data collection on violence in the Kenyan elections, was organised in Nairobi, Kenya, on the 29th of January 2018, in partnership with the Centre for Human Rights and Policy Studies (CHRIPS), and the Armed Conflict Location and events Data project. A third workshop gathering leading academic, policy and civil society actors involved in data collection of violence was held on May 24th, 2018, in London. The final project results were presented during the workshop, which also involved roundtable discussions on methods and practices in the field of data collection on violence. All three Working Papers and both Policy Briefs have now been produced and published online (the final Working Paper and final Policy Brief were published in December 2018) - in IDS OpenDocs allowing for Open Access, as well as published on the project's website. All three blogs have now also been published online on the project's website.
First Year Of Impact 2017
Sector Government, Democracy and Justice
Impact Types Policy & public services