![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
ABSTRACT
This study tests whether social media disinformation contributes to domestic terrorism within countries. I theorize that disinformation disseminated by political actors online through social media heightens political polarization within countries and that this, in turn, produces an environment where domestic terrorism is more likely to occur. I test this theory using data from more than 150 countries for the period 2000–2017. I find that propagation of disinformation through social media drives domestic terrorism. Using mediation tests I also verify that disinformation disseminated through social media increases domestic terrorism by, among other processes, enhancing political polarization within society.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Supplementary material
Supplemental data for this article can be accessed here.
Notes
1. Disinformation refers to deliberately propagated false or misleading information (Tucker et al., Citation2018).
2. While the theoretical argument and research design of the study focuses on disinformation disseminated through social media platforms, in reality a variety of online media are used to produce disinformation including blogs, websites and even online news websites. Because I look at a variety of actors that promote disinformation that have different endowments of technological or financial resources – national governments, foreign governments and domestic political parties – it makes most sense to focus on the form of online media with the lowest barriers to entry: social media.
3. Approximately 44.4% of all countries in the total sample. Note, the number of countries in the main analyses is lower due to missing data.
4. There have been a couple of empirical studies on the impact of social media and hate crimes. For example, a study by Young, Swamy, and Danks (Citation2018) determined that there is a positive relationship between misinformation and expressions of hate speech in YouTube comments and the subsequent perpetration of physical violence. Other studies draw a link between social media usage, but not misinformation, and hate crime. For example, unpublished studies by Müller and Schwarz (Citation2020, Citation2018b) found that hate crimes against immigrant victims in Germany or against Muslims living in the United States were higher in localities where social media usage and Twitter is higher.
5. As will be explained in the research design section below, the study examines the impact of the climate of social media disinformation on country-year counts of domestic terrorist attacks within the country. The study does not delve into specific terrorist or extremist actors and how they may be affected by social media disinformation.
6. This definition conforms to the operationalization scheme used for the main independent variable of the study developed by Mechkova et al. (Citation2019, p. 21). Conceptually, my use of the term political polarization mirrors closely to what Iyengar, Sood, and Lelkes (Citation2012) refer to as “affective polarization” or to what Mason (Citation2015) calls “social polarization.” Affective or social polarization can be distinguished from the traditional term polarization in that it is a mass rather than a simply elite phenomenon and it transcends single policy issues but rather is associated with political or partisan identity. See Fiorina and Abrams (Citation2008) for a full discussion of the different uses of political polarization.
7. The third process facilitated by the internet, according to Byman (Citation2018), is operational direction. In the case of the Comet Ping Pong attack, one could argue that online disinformation aided the attacker in the selection of the target, which would be an example of online misinformation contributing to terrorist operational direction.
8. The participatory nature of social media is crucial here. Social media and websites provide a participatory outlet for individuals through message boards, chat rooms and comments boards. This furthers the process of radicalization and the construction of group identity.
9. For more information about this process, see Jason Stanley interview by Illing (Citation2019). He explains, “[Politicians like Trump] flood the media zone with all kinds of bizarre nonsense … [a]nd what this does is create a complete cacophony. It’s just too much for anyone to sort out. And the result is people just say, ‘Well, who’s on my side?’ … It’s not about ideas or facts but about my side and your side … ”
10. The GTD is an event database compiled and curated by researchers at the START Center at the University of Maryland. The codebook for GTD can be found online at: https://www.start.umd.edu/gtd/.
11. Though the GTD defines terrorism as violence used by non-state actors, its coding rules do not preclude attacks by political actors that have political wings that hold governmental responsibilities (e.g., Hezbollah in Lebanon). The attacks of such actors are counted by the GTD and are included in the analysis. My expectation is that terrorism by these sorts of actors is similarly affected by social media disinformation.
12. Domestic terrorist attacks are attacks that occur within the borders of a single country and where the perpetrator and target or victims are citizens or permanent residents of the country. Transnational terrorist attacks are those where the perpetrator and the victim or targets are citizens or residents of different countries and/or where the attack transcends a national border.
13. Results available from author. Note that it is likely that these checks reproduce the main results because a majority of attacks, 80%, in the GTD are domestic.
14. Data and codebook available online at: http://digitalsocietyproject.org/.
15. Like V-Dem, the Digital Society Project uses multiple country specialists to derive measurements for its various indicators. The measures are derived through various aggregation methods, including distilling measures into ordinal scales based upon combined specialist judgements.
16. Variable names are “v2smgovdom” for domestic government dissemination of disinformation, “v2smpardom” for party dissemination of disinformation and “v2smfordom” for foreign governments dissemination disinformation in Mechkova et al. (Citation2019, p. 21). Note, in the original ordinal scale developed by Mechkova et al. (Citation2019), the scores range from 0 to 4 where a score of zero is the highest frequency of dissemination of disinformation and a 4 is the lowest frequency. To ease interpretation of results in the study, I have inverted this scale. Specifically, in the inverted scale I use in the analysis: 4 = Extremely often. Actors disseminate disinformation on all key political issues; 3 = Often. Actors disseminate disinformation on many key political issues. 2 = About half the time. Actors disseminate disinformation on some key political issues, but not others. 1 = Rarely. Actors disseminate disinformation on only a few key political issues. 0 = Never or almost never. Actors never disseminate disinformation on key political issues.
17. R = .732 for domestic government and party, .398 for domestic government and foreign government and .328 for party and foreign government.
18. “v2smonex,” “Do people consume domestic online media?” Responses: 0 = Not at all. No one consumes domestic online media; 1 = Limited. Domestic online media consumption is limited. 2 = Relatively extensive. Domestic online media consumption is common. 3 = Extensive. Almost everyone consumes domestic online media. (Mechkova et al., Citation2019).
19. “v2smgovfilprc,” “How frequently does the government censor political information (text, audio, images or video) on the Internet by filtering (blocking access to certain websites)?” Note, to ease interpretation, I invert the scale to the following: Responses: 4 = Extremely often; 3 = Often; 2 = Sometimes; 1 = Rarely; 0 = Never or almost never. (Mechkova et al., Citation2019).
20. Source: United Nations National Accounts. Available online at: https://unstats.un.org/unsd/snaama/.
21. Data and codebook are available online at: https://www.prio.org/Data/Armed-Conflict/UCDP-PRIO/.
22. These controls represent the most commonly included covariates in empirical studies of terrorist attacks. However, they do not represent an exhaustive set of potential controls. In order to address possible spuriousness produced by omitted variable bias, I take two steps. First, using Morris and LaFree (Citation2016), I identify an extended set of controls to include in estimations. These are: country Gini coefficient to measure income inequality; ethno-linguistic fractionalization; GDP growth; the protection of physical integrity rights within the country; whether or not the country is involved in an interstate war; and the urbanization rate of the country. These robustness models produce the same findings as those in the main models. I do not use them as the main models because their inclusion severely reduces the number of observations and drops several country cases due to missing-ness and list-wise deletion. Including the extended controls reduces observations by 25% and drops the number of countries in the analysis from 156 to 117. I therefore use them as robustness checks only. Appendix Table 3 presents the models with the extended controls along with a short description of the extended controls. Second, to account for any country-level features that are not captured by the controls in the main models, I also conduct country fixed effects estimations. These are presented in Appendix Table 4.
23. “v2smpolsoc,” “How would you characterize the differences of opinion on major political issues in this society?” Note, to ease interpretation, I invert the scale to the following: Responses: 4 = Serious Polarization. There are serious differences in opinions in society on almost all key political issues, which result in major clashes of views; 3 = Moderate [high] polarization. There are differences in opinions on many key political issues which result in a moderate clash of views. 2 = Medium polarization. Differences in opinions are noticeable on about half of the key political issues resulting in some clashes of views. 1 = Limited polarization. There are differences in opinions on only a few key political issues, resulting in few clashes of views. 0 = No polarization. There are differences of opinions but there is a general agreement on the direction for key political issues.
24. Results available from the author.
25. There is some evidence this is the case, though differences in social media usage between developed and developing countries are fairly modest. The average score for online media consumption is 2.1 for “developed countries” (countries in North America, Western Europe, Central and Eastern Europe and Eurasia) while it is 1.8 for “developing countries” (countries in Latin America, Africa, the Middle East, East, South and Southeast Asia).
26. Results available from author. Note that in these analyses I also include a measure of income inequality to hold constant disparate economic levels within developing countries.
27. “v2smorgtypes,” “What types of offline political action are most commonly mobilized on social media?” Answer: “v2smorgtypes_7” “Terrorism.” Average of expert score where 0=no and 1=yes.
28. Specifically, I used a statistical package for Stata designed by Hicks and Tingley (Citation2011) called “medeff” which is adapted from Imai, Keele, and Tingley (Citation2010) and Imai, Keele, and Yamamoto (Citation2010).
29. To conduct sensitivity analyses, I used the “medsens” Stata statistical package, also developed by Hicks and Tingley (Citation2011). Full results of these tests are available from the author.
30. This can be seen in ) when comparing the β coefficients for estimations where the mediator is included – the top coefficient not in parentheses – and included – the bottom coefficient in parentheses.
31. In response to the January 6 violence against the U.S. Capitol, Twitter permanently banned former U.S. President Donald Trump’s account due to his frequent dissemination of disinformation.
Additional information
Notes on contributors
James A. Piazza
James A. Piazza is Liberal Arts Professor of Political Science at The Pennsylvania State University. He is the author of over 50 articles on terrorism and political violence.