Giters

DFRLab / Dichotomies-of-Disinformation

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Dichotomies of Disinformation:
Project Overview & Codebook
(Beta)

Emerson T. Brooking, Alyssa Kann, & Max Rizzuto
Digital Forensic Research Lab, Atlantic Council
With the support of the entire DFRLab staff and their years of extensive reporting on disinformation.

Rocky Taylor Cole & Andrew Gully
Jigsaw



Visualization | Project Overview | Codebook | Dataset

Summary

Through the use of viral falsehood and digital manipulation, determined actors can destabilize civil institutions, affect social polarization, or disrupt rival militaries. As social media has proliferated, so has the use of these new stratagems. We isolate and define this phenomenon, which we describe as “political disinformation campaigns.” We also propose and test a classification system built on 150 variable options. Our intent is to establish a replicable, extensible system by which widely disparate disinformation campaigns can be categorized and compared.

Introduction

Through the deliberate seeding and manipulation of viral falsehood, a determined actor can destabilize civil institutions, affect social polarization and inflame ethnic and religious tensions, disrupt or demoralize rival militaries, sabotage foreign economies, or simply abuse the nature of digital advertising for personal financial gain. This phenomenon is broadly known as “digital disinformation” (hereafter referred to as “disinformation”). Today, at least 70 national governments or national-level political parties have invested in such information manipulation capabilities (Bradshaw & Howard 2019), as have innumerable individuals and business interests.

As the potency and prominence of disinformation has grown, several frameworks have been proposed to contextualize and understand it. These frameworks have been designed with different ends in mind. Wardle & Derakhshan (2017) take the most expansive view, identifying disinformation as just one contributor, alongside mis- and mal-information, to a state of “information disorder.”

Other frameworks emphasize narrative analysis and deconstruction. Pamment et al. (2018) provide a diagnostic of “influence activities” conducted by foreign powers, of which disinformation is a subset. They divide each discrete campaign into rhetorical strategies (positive, negative or oblique); techniques (forging and leaking to deceptive identities); and influence stratagems (black propaganda to polarization to flooding). By applying this framework, policymakers can formulate more effective counter-messages.

Still other frameworks offer greater granularity and narrower ends. Bradshaw & Howard (2017; 2018; 2019) have created an evolving schema of organizational forms, account types, and general strategies that describe both disinformation campaigns and related information manipulation techniques. This system helps describe and differentiate national approaches to the digital commons. Martin & Shapiro (2019) present a detailed classification system for online foreign influence efforts by which attacker characteristics, tactics, and rhetorical tactics, and platform choice can be assessed and compared.

Even amid this abundance of studies and frameworks, however, critical gaps remain. By casting a wide net (e.g. “influence activities”), some frameworks fail to distinguish disinformation from state propaganda or illicit financial pressure, which have much longer histories. By placing heavy emphasis on rhetorical and narrative classification, many frameworks also elude useful application to non-Western nations, in which foreign researchers may be unfamiliar with the language and local political context. Because many frameworks focus exclusively on the activities of adversarial powers, they do not address the challenge of classifying disinformation campaigns of domestic origin.

We have created and applied a new framework for disinformation classification. This framework is intended to be replicable and extensible, able to describe and compare widely disparate campaigns. This short paper proceeds in three parts: a presentation of definitions, an explanation of case selection and methodology, and a discussion of initial project observations and areas for further research. The project codebook itself is contained in Appendix A.

Definition

We define disinformation as “false or misleading information, spread with the intention to deceive” (Nimmo 2018). This definition captures both falsehood and deception by means of decontextualization. This definition also emphasizes intentionality. Contrary to some competing definitions, we make no claims with relation to the purpose of the disinformation.

We define political disinformation as “disinformation with a political or politically adjacent end.” This definition captures disinformation spread in the course of an election, protest, or military operation. It also captures the widespread phenomenon of political “clickbait” disseminated for personal financial gain.

Finally, we define a political disinformation campaign as “political disinformation that demonstrates both coordination and a discrete objective.” As objectives change over time, it becomes possible to delineate one political disinformation campaign from another. Note that while a discrete objective will always exist, it may not always be discernible.

Although not explicated in the definition, political disinformation campaigns almost always evidence the inauthentic amplification of content: what Heller et al. (2019) call “political astroturfing” and what Facebook identifies as “coordinated inauthentic behavior” (Gleicher 2018). Although inauthentic amplification is not “disinformation” per se, it is intended to create an illusion of consensus or popularity, and is therefore inextricably linked to political disinformation campaigns. Selected cases do not include instances of political misinformation, in which evidence of coordination or intent is absent.

Case Selection and Methodology

We have established two criteria for case selection. First, a case must meet the definition of a political disinformation campaign. This is done by means of inter-coder agreement, with guidance from the principle investigator.

Second, the case must be drawn from a trustworthy secondary source. By coding the work and conclusions of other researchers, we can mitigate—although not eliminate—appearances of an opaque or arbitrary classification process. This is especially important because of the reputational risk that comes with false attribution of political disinformation campaigns.

In this initial project phase, we used DFRLab publications for all coded cases. This ensured source consistency and certainty regarding the rigor of research processes. As the project develops further, we envision the inclusion of cases from trusted publications and research institutions. This “trusted” status will be established by means of the nine NewsGuard credibility and transparency standards, or a similarly well-regarded system of source evaluation.

Following case selection, we built a system of classification. This framework was developed in an iterative process. Coders used test cases to expand upon definitions and refine the taxonomy. Cases were classified by two coders and inter-coder reliability was tracked to ensure consistent application of categories and definitions. Variables that were not consistently applied were refined or removed.

Ultimately, we created a system with 76 binary variables, 35 text descriptors, and 40 quantitative variables. This system classifies political disinformation campaigns through six categories: Target, Platforms, Content, Methods, Attribution, and Intent. This order reflects the process by which the DFRLab identifies and evaluates suspected campaigns.

  • Target describes the national or supranational characteristics of the target, to include its political or social strata.
  • Platform describes the medium through which the disinformation is conveyed, to include open web, social media, and messaging services.
  • Content describes the language and topic of the disinformation.
  • Methods describes the tactics and narrative techniques that are used to disseminate the disinformation.
  • Attribution describes the national or supranational characteristics of the disinformant, to include its political or social strata, as well as the confidence of the attribution.
  • Intent describes the inferred purpose of the political disinformation campaign.

The question of attribution required particular attention. The system delineates between two levels of attribution to state actors. The first level is a declaration by a social media platform or a trusted government entity, which have access to signals intelligence and other publicly unavailable information. The second level is attribution to a state proxy, based either on a credible third-party assessment or the informed evaluation of a DFRLab researcher.

Such DFRLab evaluation is based on a close study of open-source evidence, including how directly a campaign can be tied to specific, named actors. The more compelling and clear these connections, the higher the confidence of this assessment. In many cases, no affirmative attribution can be made.

We proceeded to code 60 cases through application of this framework. Cases were classified by two coders and inter-coder reliability was tracked. In addition to a close reading of the associated article, coders also followed hyperlinks to open-source evidence (when available) to determine any additional language or platform uses. Coder disagreement was resolved by the principle investigator.

Initial Observations & Areas for Further Research

Many of the most striking initial observations in the dataset do not regard the character of individual campaigns, but rather the process of the coding itself. Because DFRLab case studies are reflective of the broader field of disinformation studies, this process is revealing about the state of the discipline as a whole.

By a significant margin, the most targeted nation in the dataset is the United States (eight). The second-most targeted is Indonesia (four), followed by Venezuela and Mexico (three each). The most targeted region as a whole is the Middle East (three). A quarter of campaigns (15) had more than one target nation or region recorded. The electorate was the target in an overwhelming number of cases (52), followed by civilian government (11) and political party (10). There were 13 domestic cases, all of which targeted categories that included the electorate. At least one concurrent event was coded in the majority of cases (44). As a whole, inter-coder agreement on target binaries was 91 percent.

The most popular disinformation platform in the dataset is Facebook (39), followed by Twitter (30). One-third of cases made use of “junk news” websites (20). As a whole, inter-coder agreement on platform binaries was 96 percent: the highest across the project.

Regarding the content of political disinformation campaigns, the most common language was English (31), followed by Spanish (16) and Russian (9). The overwhelming topic focus was on government (47), followed by political parties (29). The third most common topic was racial, ethnic, religious, or sexual minorities (17), reflecting the frequent marginalization of vulnerable groups in political disinformation campaigns. As a whole, inter-coder agreement on content binaries was 82 percent.

The technical and rhetorical methods of disinformation were the most difficult to assess in a consistent fashion. The most common technical tactic was the use of sockpuppets (32), followed by botnets (18) and brigading (15). Constructive narratives were by far the most common, both in order to astroturf (36) and activate real supporters of a movement (21). Destructive narratives were also common, both for the purpose of discrediting (28) and suppression of political opposition (5). As a whole, inter-coder agreement on methods binaries was 87 percent.

In attribution of political disinformation campaigns, two clear disinformants stood out: Russia (9) and Iran (7). Campaigns were most often attributed to government entities, either via direct attribution (10), or via indirect attribution to proxies working on the government’s behalf (16). A number of business entities (7) also participated in political disinformation. Although the average campaign length of a case in our dataset was 2.5 years, the average length of a Russian nation of origin campaign was 6.2 years, and 5.2 years for Iran. The third most often attributed power was Venezuela (3). As a whole, inter-coder agreement on attribution binaries was 92 percent.

Finally, regarding the intent of political disinformation campaigns, two-thirds sought to alter the target’s civil institutions (40). Nearly as many carried a discernible social objective (36). Few were conducted in direct support of military campaigns (6), and just one had an obvious economic objective. As a whole, inter-coder agreement on intent binaries was 84 percent.

Given a popular focus on Russia as the world’s primary disinformant, it was surprising to see Iran rank nearly as high in the dataset. It was also surprising, given a widespread obsession with “bots” as the source of disinformation, that less than one-third of cases employed obvious botnets. More predictable was the high number of destructive narratives, spread with the intent of discrediting particular individuals or institutions. Such negative goals remain the principle use of political disinformation campaigns.

There are clear limitations to this project, however. As a consequence of our definition, the number of political disinformation campaigns is potentially inexhaustible. Because of our case selection system, political disinformation campaigns may also overlap. This means that any comparative conclusion from our dataset (e.g. “Russia conducts more disinformation operations than Iran”) cannot be sufficiently substantiated. It also means, because less reporting is available on cases that target the Global South, that certain nations are over- or underrepresented in this database.

The intention of this project has been to construct and evaluate a new schema for the classification of political disinformation campaigns that is both replicable and extensible. Ideally, this system might be applied by other academic institutions and journalistic outlets.

The ultimate goal of this project is to establish a large database of political disinformation campaigns. Researchers, policymakers, journalists, and concerned citizens might analyze this database for insights into the aggregate functions of disinformation campaigns. This will assist in the effort to identify and isolate these campaigns before they take form.


Appendix A: Codebook

Meta | Target | Platform | Content | Methods | Attribution | Intent

Taxonomy and Definitions

The indentation may be a bit off, but now the hierarchy is: I, 1, A, a, i. This helps get us shorthand unique IDs for each of the variables. Variables are in plain text; definitions are italicized. All variables are marked with one of these “data types”:

  • binary
  • decimal (ie, 2.4)
  • integer (ie, 2 or -4)
  • free text
  • country (all countries are options in a drop-down)
  • language (all languages are options in a drop-down)
  • bloc (all regional blocs are options in a drop-down)
  • year (relevant years are options in a drop-down)
  • month (all months are option in a drop-down)
  • range (of years -- pre-determined from data source)

  • Meta

    • Serial Number (integer)
    • Campaign ID (<STARTING YEAR> <ATTACKER> against <TARGET>). This is intended to aid categorization and reconciliation of overlapping campaigns.
    • Long Description (free text). Summary of the campaign in one to two sentences.
    • Dates
      • Start
        • Starting Year (year). Starting year of when the campaign first began. The starting year can be inferred from context, or be explicitly stated in text.
        • Starting Month (month).
        • Certainty Start (binary). Input 1 when the source document clearly indicates that the campaign began in this time frame.
      • End
        • Ending Year (year)
        • Ending Month (month). In the event of a takedown, the ending month is the month of the takedown (unless it is demarcated that the assets stopped being active prior to the takedown date). For cases that are still occuring at the time of coding, the ending year and month should be left blank.
        • Certainty End (binary). Input 1 when the source document clearly indicates that the campaign ended in this time frame.
    • Sources
      • Year of Report (year). Year of Citation 1.
      • Month of Report (month)
      • Citation 1 (free text). Chicago style.
      • Link 1 (free text)
      • Citation 2 (free text). Chicago style.
      • Link 2 (free text)
    • Notes

  1. Target

    1. Primary Target

      1. Nation of Origin (Country) This should be filled out even when the target is not a nation. When a campaign targets a non-state political actor, the nation of origin of that non-state political actor is filled in this field, if that information is available. Distinguishable territories are nations.
      2. Regional Bloc (Bloc). When a single nation of origin cannot be determined.
      3. Other (free text)
      4. Notes (free text)

    2. Target Category. Categories are not mutually exclusive. All relevant categories can be added.

      1. Government. If a determination cannot be made between Civilian and Military categories, the Civilian category takes precedence.
        1. Civilian (binary). The governing body and functions of a state, including national leaders, institutions, and non-military departments and agencies. Includes incumbent politicians running for re-election.
        2. Military (binary). Military departments and agencies which enjoy the sanctioned use of force.
      2. Political Party (binary). Organized competitors for political power who can obtain or wield power directly. Includes politicians currently in office, as well as non-incumbent politicians running for office who are associated with a political party. Can also be an individual working for a party.
      3. Non-State Political Actor (binary). Organized competitors for political power who can obtain or wield power, even if indirectly; not necessarily enfranchised. Non-state political actors are formally organized, coordinated, and cohesive. e.g. Greenpeace, the NRA, or the KKK.
      4. Business (binary). Includes groups that contract out to the government, individuals looking for financial gain, and mercenaries.
      5. Influential Individuals (binary). ndividuals who are influential but who do not belong to a ruling government coalition. Includes groups of individuals who are not formally organized but work together. e.g. journalists, former politicians, or organized 4channers. For individuals who operate their own charitable foundations (and thus could be placed in Non-State Political Actor), coding depends on whether or not the disinformation is foremost targeting the individual, their foundation, or both.
      6. Electorate (binary). The enfranchised population in a specific country or within a demarcated boundary.
      7. Racial, Ethnic, Religious, or Sexual Identity Group (binary). A specific minority/majority group.
      8. Other (free text)
      9. Notes (free text)

    3. Quantitative Measures. If nation of origin is filled out, this category should be filled out.

      1. Political Stability. Use the Estimate figure of the WGI data, “Political Stability and Absence of Violence/Terrorism.” For 2019 and onwards, use most recently available data.
        1. Political Stability Data. Year prior to campaign start.
          1. Political Stability (decimal)
          2. Data Year (year)
        2. Political Stability Data. Campaign start year.
          1. Political Stability (decimal)
          2. Data Year (year)
        3. Political Stability Data. Year after campaign start.
          1. Political Stability (decimal)
          2. Data Year (year)
      2. Refugee % Change
        1. Refugee % Change (decimal). Use UNHCR data, table 6, “Annual rate of change of refugee stock.”
        2. Data Years (range). Use “2010-2015” for 2015, to show historical progression. For 2018 and onwards, use most recently available data.
      3. Voice and Accountability. Use the Estimate figure of the WGI data, “Voice and Accountability.” For 2019 and onwards, use most recently available data.
        1. Voice and Accountability Data. Year prior to campaign start.
          1. Voice and Accountability (decimal)
          2. Data Year (year)
        2. Voice and Accountability Data. Campaign start year.
          1. Voice and Accountability (decimal)
          2. Data Year (year)
        3. Voice and Accountability Data. Year after campaign start.
          1. Voice and Accountability (decimal)
          2. Data Year (year)
      4. Internet Freedom. Use FH data. Input number only, not ranking out of 100.
        1. Internet Freedom. Year prior to campaign start.
          1. Internet Freedom (integer)
          2. Data Year (year)
        2. Internet Freedom. Campaign start year.
          1. Internet Freedom (integer)
          2. Data Year (year)
        3. Internet Freedom. Year after campaign start.
          1. Internet Freedom (integer)
          2. Data Year (year)

    4. Concurrent Events

      1. Inter-state war (binary). Threshold is 1,000 conflict deaths. Use COW data for 2007 and before. 2008 and after, supplement with research.
      2. Extra-state war (binary). Threshold is 1,000 conflict deaths.Use COW data.
      3. Intra-state wars (binary). Threshold is 1,000 conflict deaths. Use COW data for 2007 and before. 2008 and after, supplement with research.
      4. Non-state war (binary). War in non-state territory or across state borders. Threshold is 1,000 conflict deaths. Use COW data for 2007 and before. 2008 and after, supplement with research.
      5. Federal Election (binary).
      6. State Election (binary). Includes elections at province, municipality, administrative region, department, prefecture, and local levels.
      7. Other (free text)
      8. Notes (free text)

    5. Secondary Target. This should rarely be used.

      1. Nation of Origin (country)
      2. Regional Bloc (bloc)
      3. Other (free text)
      4. Notes (free text)

    6. Tertiary Target. This should rarely be used. When there are more than three targets, define a regional bloc or code as different cases.

      1. Nation of Origin (country)
      2. Regional Bloc (bloc)
      3. Other (free text)
      4. Notes (free text)

  2. Platforms

    1. Open Web

      1. State Media (binary). Includes “state-adjacent” media, operated by government proxies or otherwise beholden to the state.
      2. Independent Media (binary)
        3. Media institutions that are not beholden to the government and can be reasonably assessed to score > 60 by the NewsGuard rating process.
      3. "Junk News" Websites (binary). A website that trafficks in deceptive headlines, fails to correct errors, avoids disclosure of funding sources, and avoids labeling advertisements. One that can be reasonably assessed to score < 60 by the NewsGuard rating process.

    2. Social Media. Social media accounts created by the disinformants for deceptive purposes.

      1. Facebook (binary)
      2. Instagram (binary)
      3. Twitter (binary)
      4. Youtube (binary)
      5. LinkedIn (binary)
      6. Reddit (binary)
      7. VK (binary)
      8. Forum Board (binary)
      9. Other (free text)

    3. Messaging Platforms

      1. WhatsApp (binary)
      2. Telegram (binary)
      3. Signal (binary)
      4. Line (binary)
      5. WeChat (binary)
      6. SMS (binary)
      7. Other (free text)

    4. Advertisement (binary). Advertisements purchased by disinformants to disseminate a message of disinformation. Includes ads on social media and the open web.

    5. Email (binary)

    6. Other (free text)

    7. Notes (free text). Add more detailed information on social media platforms used (retweets, groups, etc), if applicable. Add specific numbers of social media metrics, if applicable.

  3. Content

    1. Language (language). The language of the disinformation. In coding, separate languages with a comma; don’t use “and.”

    2. Topics. Subject evident in the campaign.

      1. Government (binary). Includes international governing bodies.
      2. Military (binary)
      3. Political Party (binary)
      4. Elections (binary)
      5. Non-State Political Actor (binary)
      6. Business (binary)
      7. Influential Individuals (binary)
      8. Racial, Ethnic, Religious, or Sexual Identity Group (binary)
      9. Terrorism (binary)
      10. Immigration (binary)
      11. Economic Issue (binary)
      12. Other (free text)

    3. Notes (free text)

  4. Methods

    1. Tactics

      1. Brigading (binary). Patriotic trolls or organic coordination in which disinformants seemingly operate under their real identities. A concentrated effort by one online group to manipulate another, e.g. through mass-commenting a certain message.
      2. Sockpuppets (binary). Inauthentic social media accounts used for the purpose of deception which evidence a high likelihood of human operation. This includes catfishing and other highly tailored operations conducted under inauthentic personas.
      3. Botnets (binary). Inauthentic social media accounts used for the purpose of deception which evidence a high likelihood of automation. These accounts evidence no sustained human intervention beyond the effort necessary to program them initially. They often form large networks for the purpose of inauthentic amplification. This includes both fresh and repurposed accounts.
      4. Search Engine Manipulation (binary). Undermining search engine optimization techniques with the intention of creating an inorganic correlation of search queries and results. Often realized by way of cooperative efforts by online communities. e.g. “Google Bombing.” May also include typosquatting with the intention to mislead or redirect to another URL.
      5. Hacking
        1. DDoS (binary). Distributed denial-of-service. Malicious attempt to disrupt server traffic. In the context of political disinformation campaigns, this is intended to make it more difficult for the target to launch an effective counter-messaging effort.
        2. Data Exfiltration (binary). The unauthorized movement of data. In the context of political disinformation campaigns, this is the acquisition of sensitive information through spearphishing or similar techniques that can be subsequently released by the disinformant to boost their messaging effort.
      6. Deceptive Content Manipulation (binary). Any content that has been deceptively edited by use of Photoshop or similar software. This includes the deceptive co-option and re-use of extant media branding and style guides. This does not include the use of deep learning processes.
        1. Deep learning processes (binary). Augmented or fabricated content produced using deep learning processes. Includes “deep fakes,” “deep voice,” and textual generation.
      7. Other (free text)
      8. Notes (free text)

    2. Narrative Techniques

      1. Constructive
        1. Activate (binary). Bandwagon, pander, ignite. e.g., “If you love Mr. Trump, RT this.”
        2. Astroturf (binary). Artificial consensus-building, inflation, or amplification. Also called a “Potemkin Village.” e.g., “The #1 trending hashtag can’t be wrong.”
      2. Destructive
        1. Suppress (binary). Harass, intimidate, exhaust. Often targets influential individuals.
        2. Discredit (binary). Libel, leak, tarnish. Often targets government, political parties, elections, or other institutions.
      3. Oblique
        1. Troll (binary). Confusion by way of discourse infiltration and targeted distraction. Conscious efforts by disinformants to derail political movements through tailored engagement.
        2. Flood (binary). Confusion by way of hashtag invasion and mass noise generation. The hijacking of an online political movement through appropriation of an existing hashtag and addition of large quantity of unrelated material.
      4. Notes

    3. Notes

  5. Attribution

    1. Primary Disinformant

      1. Nation of Origin (Country) This should be filled out even when the attacker is not a national government. When a campaign is run by a non-state political actor, the nation of origin of that non-state political actor is filled in this field, if that information is available. Likewise, this should be filled out if the preponderance of attacker activity originates from within a single nation. Distinguishable territories are nations.
      2. Regional Bloc (Bloc). When a single nation of origin cannot be filled out.
      3. Other (free text)
      4. Notes (free text)

    2. Disinformant Category. Categories are not mutually exclusive. All relevant categories can be added.

      1. Government.
        1. Direct Attribution (binary). Public, definitive attribution to a national government by a social media platform or trusted government entity. These entities have access to signals intelligence and other publicly unavailable information.
        2. Proxy/Inferred Attribution (binary). Informed attribution to a government or government-adjacent proxy in which definitive proof is absent. Such attribution is based on open-source data and inference. This includes attribution to political parties, non-state political actors, businesses, and influential individuals who are suspected to be working at the government’s direction.
      2. Political Party (binary). Organized competitors for political power who can obtain or wield power directly. Includes politicians currently in office, as well as non-incumbent politicians running for office who are associated with a political party. Can also be an individual working for a party.
      3. Non-State Political Actor (binary). Organized competitors for political power who can obtain or wield power, even if indirectly; not necessarily enfranchised. Non-state political actors are formally organized, coordinated, and cohesive. e.g. Greenpeace, the NRA, or the KKK.
      4. Business (binary). Includes groups that contract out to the government, individuals looking for financial gain, and mercenaries.
      5. Influential Individuals (binary). Individuals who are influential but who do not belong to a ruling government coalition. Includes groups of individuals who are not formally organized but work together. e.g. journalists, former politicians, or organized 4channers. For individuals who operate their own charitable foundations (and thus could be placed in Non-State Political Actor), coding depends on whether or not the disinformation is foremost targeting the individual, their foundation, or both.
      6. Electorate (binary). The enfranchised population in a specific country or within a demarcated boundary.
      7. Racial, Ethnic, Religious, or Sexual Identity Group (binary). A specific minority/majority group.
      8. Other (free text)
      9. Notes (free text)

    3. Quantitative Measures. If nation of origin is filled out, this category should be filled out.

      1. Political Stability. Use the Estimate figure of the WGI data, “Political Stability and Absence of Violence/Terrorism.” For 2019 and onwards, use most recently available data.
        1. Political Stability Data. Year prior to campaign start.
          1. Political Stability (decimal)
          2. Data Year (year)
        2. Political Stability Data. Campaign start year.
          1. Political Stability (decimal)
          2. Data Year (year)
        3. Political Stability Data. Year after campaign start.
          1. Political Stability (decimal)
          2. Data Year (year)
      2. Refugee % Change
        1. Refugee % Change (decimal). Use UNHCR data, table 6, “Annual rate of change of refugee stock.”
        2. Data Years (range). Use “2010-2015” for 2015, to show historical progression. For 2018 and onwards, use most recently available data.
      3. Voice and Accountability. Use the Estimate figure of the WGI data, “Voice and Accountability.” For 2019 and onwards, use most recently available data.
        1. Voice and Accountability Data. Year prior to campaign start.
          1. Voice and Accountability (decimal)
          2. Data Year (year)
        2. Voice and Accountability Data. Campaign start year.
          1. Voice and Accountability (decimal)
          2. Data Year (year)
        3. Voice and Accountability Data. Year after campaign start.
          1. Voice and Accountability (decimal)
          2. Data Year (year)
      4. Internet Freedom. Use FH data. Input number only, not ranking out of 100.
        1. Internet Freedom. Year prior to campaign start.
          1. Internet Freedom (integer)
          2. Data Year (year)
        2. Internet Freedom. Campaign start year.
          1. Internet Freedom (integer)
          2. Data Year (year)
        3. Internet Freedom. Year after campaign start.
          1. Internet Freedom (integer)
          2. Data Year (year)

    4. Concurrent Events

      1. Inter-state war (binary). Threshold is 1,000 conflict deaths. Use COW data for 2007 and before. 2008 and after, supplement with research.
      2. Extra-state war (binary). Threshold is 1,000 conflict deaths.Use COW data.
      3. Intra-state wars (binary). Threshold is 1,000 conflict deaths. Use COW data for 2007 and before. 2008 and after, supplement with research.
      4. Non-state war (binary). War in non-state territory or across state borders. Threshold is 1,000 conflict deaths. Use COW data for 2007 and before. 2008 and after, supplement with research.
      5. Federal Election (binary). National level election. Supra-national bodies are included.
      6. State Election (binary). Includes elections at province, municipality, administrative region, department, prefecture, and local levels.
      7. Other (free text)
      8. Notes (free text)

    5. Secondary Disinformant. This should rarely be used.

      1. Nation of Origin (country)
      2. Regional Bloc (bloc)
      3. Other (free text)
      4. Notes (free text)

    6. Tertiary Disinformant. This should rarely be used. When there are more than three disinformants, code as different cases.

      1. Nation of Origin (country)
      2. Regional Bloc (bloc)
      3. Other (free text)
      4. Notes (free text)

  6. Intent. This is the intent of the primary disinformant and any other disinformants coded.

    1. Object (free text). One or two succinct sentences.

    2. Category. A clearly definable intent may not exist.

      1. Civil (binary). To include electoral interference, policy change.
      2. Social (binary). To include marginalization of majority/minority groups and general social fissure.
      3. Economic (binary). To include suppression of economic activity, destruction of capital.
      4. Military (binary). To include complement to offensive military campaign, or information paralysis of an adversary’s military institutions.

    3. Notes (free text)

Dataset

File Format Download Size
.xlsx 75 KB
.csv 127 KB
.JSON --
.R --

References

Bradshaw, Samantha, and Philip Howard. "Troops, trolls and troublemakers: A global inventory of organized social media manipulation." (2017).

------- "Challenging truth and trust: A global inventory of organized social media manipulation." The Computational Propaganda Project (2018).

------- " The Global Disinformation Order: 2019 Global Inventory of Organised Social Media Manipulation." The Computational Propaganda Project (2019).

Gleicher, Nathaniel, “Coordinated Inauthentic Behavior Explained,” Facebook, December 6, 2018.

Keller, Franziska B., David Schoch, Sebastian Stier, and JungHwan Yang. "Political Astroturfing on Twitter: How to Coordinate a Disinformation Campaign." Political Communication (2019): 1-25.

Martin, Diego A., and Jacob N. Shapiro. "Trends in Online Foreign Influence Efforts." (2019).

Nimmo, Ben, and Graham Brookie. "Fake News: Defining and Defeating," DFRLab, January 19, 2018.

Pamment, James, Howard Nothhaft, and Alicia Fjällhed. "Countering Information Influence Activities: A Handbook for Communicators." (2018).

Wardle, Claire, and Hossein Derakhshan. "Information Disorder: Toward an interdisciplinary framework for research and policy making." Council of Europe Report 27 (2017).



   

About