Terrorism as group threat

“It is the events seemingly loaded with great collective significance that are the focal points of the public discussion. […] When this public discussion takes the form of a denunciation of the subordinate racial group, signifying that it is unfit and a threat, the discussion becomes particularly potent in shaping the sense of social position” Blumer (1958): 6

Terrorism as group threat

• Terrorist attacks pose a threat to individual and collective safety
• And strengthen intolerance and prejudice
• H1: Ethno-religious hostility increases after terror attacks
• However, effect tends to be short-lived (Hopkins 2010; Legewie 2013)
• H2: The effect of H1 levels off over time

Prior research: Survey data

• Natural experiments find mixed results
• More negative attitudes after a terror attack
  • E. g.: Hopkins 2010; Legewie 2013, Böhmelt et al. 2019; Finseraas et al. 2011; Nussio, Bove, and Steele 2019
• Little or no effect
  • E. g.: Brouard et al. 2018; Finseraas et al. 2011; Larsen et al. 2019
• Methodological concerns and social desirability bias

Prior research: Social media data

• Tweets sympathetic yet nationalistic after attack (Fischer-Preßler et al. 2019)
• YouTube comments more hostile (Czymara et al. 2023)
• Tweets became more similar to AfD language (Giavazzi et al 2023)

General and specific components of threat

• Attacks can shape attitudes toward groups that are often associated with the attacks
• H3a: Islamist attacks have the strongest effect on Tweets about Muslims
• However, terrorism sometimes blamed on immigration in general (Helbling & Meierrieks 2020)
• H3b: There is a spillover effect onto Tweets about migration in general

Mechanism

• Attitudinal change vs. compositional change
• Group threat theory implies users update their views
• H4: Within users’ hostility increases after terror attacks

Overview of attacks

Source: https://www.start.umd.edu/gtd/search/

Search string

• Goal: Identify Tweets that deal with migration related topics
• Should include Muslims (much debated migrant group)
• But no direct search for insults or derogatory terms, nor attack specific hashtags
• Language-specific search string, English version:

 [1] "migration"   "immigration" "migrant"     "immigrant"   "foreigner"  
 [6] "migrants"    "immigrants"  "foreigners"  "refugee"     "refugees"   
[11] "asylum"      "islam"       "muslim"      "muslims"    

Distribution of Tweets before and after attacks

Measuring ethno-religious hostility

• Relying on of Google Jigsaw’s Conversation AI’s Perspective
• “Perspective is trained to recognize a variety of attributes (e.g. whether a comment is toxic, threatening, insulting, off-topic, etc.) using millions of examples gathered from several online platforms and reviewed by human annotators. Probability scores represent a probability, with a value between 0 and 1. A higher score indicates a greater likelihood that a reader would perceive the comment as containing the given attribute.” (Google)

Measuring ethno-religious hostility

• Perspective’s IDENTITY_ATTACK attribute, which measures “Negative or hateful comments targeting someone because of their identity.” (Google)
• In this context, I assume identity proxies ethnicity, race, or religion
• Based on convolutional neural networks (Dixon 2018), details of the algorithm are proprietary

Validation

• Annotated a random sample of Tweets, stratified by hostility level and case
• Almost perfectly selected hostile Tweets only: Precision of 98 percent (\(\frac{true positive}{true positives + false positives}\))
• High ability to detect hostility: Recall of 82 percent
  (\(\frac{true positive}{true positives + false negatives}\))

Statistical model

• Dependent variable: Probability to be ethno-religiously hostile (ranging from 0 to 100)
• Predictor: Posted either in the week before or after the attack (binary variable)
• Running variable that counts the time before and after the attack (in minutes) to measure trends
• Interrupted time series: \(p(hostility)=\beta_0+\beta_1attack+\beta_2time\_since\_attack +\) \(\beta_3attack*time\_since\_attack+\beta_4case+\epsilon\)
• Cluster-robust standard errors at the user level

Results (H1 & H2)

Putting effect size into perspective

• Hourly trend after attack: -0.06 percentage points
• Daily trend after attack: -1.443 pp
• The hostility level has returned to its prior level after 6.845 days

Variation across attacks

• Allow both the attack effect and the trends to vary across cases
• Three-way interaction model: \(p(hostility)=\beta_0+\beta_1attack+\beta_2time\_since\_attack+\beta_3case +\) \([twoway\_interactions]+\beta_4attack*time\_since\_attack*case+\epsilon\)

Variation across attacks

Non-linear trends

Without Tweets about Muslims or Islam (H3a & H3b)

• 909841 Tweets do not mention Muslims or Islam at all
• Immediate attack effect: 8.156 pp (\(p<\) 0.001) (compared to effect in pooled model: 9.877 pp)
• Pre-attack trend: -0.001 pp per hour (\(p=\) 0.19)
• Post-attack trend: -0.047 pp per hour (\(p<\) 0.001)

Intra-individual change (H4)

• Fixed effects model
• Analyzing a subset of accounts that Tweeted at least once before and after an attack
• Somewhat different sample

Intra-individual change (H4)

• Immediate attack effect: 3.594 pp (\(p<\) 0.001) (compared to effect in pooled model: 9.877 pp)
• Pre-attack trend: 0 pp per hour (\(p=\) 0.088)
• Post-attack trend: -0.019 pp per hour (\(p<\) 0.001)

Robustness checks

• Binary outcomes (75% & 90% threshold) and logistic regression
• Removing top 1% of highly active accounts (bots?)
• Only before period, cut in half: no effect
• Random treatment timing: no effect

Conclusion

• Islamist terrorism fuels online hate across cases and countries
• Higher hostility levels after attacks
• Due to individuals becoming more hostile
• … and due to new people entering the debates
• Limitation: deletion of hate speech (results overly conservative)

Finally

• Just published in the Journal of Computational Social Science at https://doi.org/10.1007/s42001-024-00272-9 (open access)
• Replication material available on the Open Science Framework
• More information: czymara.com