2019 GLOBAL INVENTORY OF ORGANISED SOCIAL MEDIA MANIPULATION For the past three years, our three-stage methodology has Persian, Polish, Portuguese, Russian, and Spanish. We also allowed us to successfully capture a wide range of public worked with BBC monitoring1 who provided an additional portal documents that shed light on the organized manipulation for collecting and aggregating high-quality news and information campaigns globally. There are almost certainly cyber troop on cyber troop activity, as well as translation services for news operations that have not been publicly documented, and we articles for Bosnia, Croatia, Georgia, Kazakhstan, Kyrgyzstan, have already seen these cases grow over time. While this report Malaysia, North Macedonia, Taiwan, Tajikistan, Turkmenistan, in no way is intended to provide a complete picture of how state Uzbekistan. We relied on English-language-only reporting for: actors are operating in this space, we can begin to build a bigger Armenia, Azerbaijan, Cambodia, China, Czech Republic, Eritrea, picture by piecing together public information. The country- Ethiopia, Hungary, Israel, Moldova, Myanmar, Netherlands, specific profiles and a full list of news items and secondary North Korea, Pakistan, Philippines, Serbia, South Korea, Sri literature sources can be found on the 2019 report homepage. Lanka, Thailand, Turkey, and Vietnam. Content analysis is an established research method in After conducting a content analysis, a team of research communication and media studies (Herring 2009). It has been assistants completed a secondary literature review to provide used to help understand how the Internet and social media an in-depth profile of cyber troop activity in a specific country interact with political action, regime transformation, and context. These case studies drew from the data collected in digital control (Bradshaw and Howard 2018a, 2017b; Edwards, the content analysis, as well as an in-depth secondary literature Howard, and Joyce 2013; Joyce, Antonio, and Howard 2013; review, where case study authors searched for other high- Strange et al. 2013). This qualitative content analysis was quality open source information about cyber troop activity. This conducted to understand the range of state actors who actively involved looking for government reports, think tank papers, use social media to manipulate public opinion, as well as their academic and scholarly studies, and research conducted by capacity, strategies, and resources. We modelled our content civil society organizations. A complete archive of the news analysis after last year’s report, using purposive sampling to sources and secondary literature used in this report can be build a coded spreadsheet of specific variables that appear found in an online Zotero database. We hope this public library in news articles. The following keywords were selected and will help inform future research. used in combination for our search: bot; Cambridge Analytica; After completing a qualitative content analysis and secondary disinformation; Facebook; fake account; information warfare; literature review, research assistants synthesized the findings Instagram; military; misinformation; propaganda; psychological into short country case studies. The case studies provide more operations; psyops; social media; sock puppet; troll; Twitter; information about instances of computational propaganda we WhatsApp; YouTube. identified in the content analysis, as well as detailed information There are two major limitations to conducting our qualitative about the specific country context and media environment in content analyses: media bias and language. To help mitigate which social media manipulations are taking place. In addition bias, we used LexisNexis and the top three search engine to the content analysis and secondary literature review, we providers – Google, Yahoo! and Bing – which provided hits to completed a case study for 84% of the countries, which can be a variety of professional, local, and amateur news sources. To online in a data supplement alongside the report. ensure that only high-quality news sources were being used to Finally, the last step of our research methodology – build our dataset, each article was given a credibility score using consultations with experts – allowed us to peer review the a three-point scale. Articles ranked at one came from major, case studies, as well as get feedback on the quality of English professionally branded news organizations. Articles ranked at and local-language news reporting and secondary literature two came from smaller professional news organizations, local we found and discuss additional resources and citations in news organizations, or expert commentary and professional alternative languages with native speakers. Experts were asked blogs. Articles ranked at three came from content farms, or to review the case studies drafted by research assistants, and personal or hyper-partisan blogs. These articles were removed (1) fact-check the information and data for accuracy; (2) provide from the sample. additional citations to open source material; and (3) provide Language was a second limitation to conducting our qualitative general feedback on the reliability of the data. In the cases of content analysis. For this year’s global inventory, we were able Poland, Sri Lanka, Taiwan, Tunisia, and Ukraine, we consulted to draw upon news articles and secondary resources written experts on the data collected from the content analysis and in Arabic, English, French, German, Greek, Hungarian, Italian, literature review. 1 https://monitoring.bbc.co.uk/ [ 8 ]