Embedded Files
Is there a systematic bias in how the U.S. media covers terrorism? Many have speculated that attacks by Muslims are covered in a very different way than attacks by right-wing extremists. Some research has shown that Muslim attacks receive a disproportionate amount of media coverage. However, that says nothing about the content of that coverage. We use cutting edge statistical techniques to discern if there is a difference in media coverage.
Background and Research Design
There is research that shows that the U.S. media disproportionately covers terrorism (or attempted terrorism) committed by Muslims. We are interested in the specific content of the coverage. For instance, in terrorist actions (classified as such by independent research centers) committed by right-wing extremists are often not even referred to as terrorism in the coverage. This would exacerbate the disproportionate coverage issue - not only are attacks by Muslims covered much more than is warranted given the actual numbers of attacks, the language in that coverage may also be biased. We attempt to test this using structural topic models (STMs). These models take text and find common word associations and group them, through an algorithm into topics (i.e. "missile" could be associated with a topic of "war"). These topics can be thought of as ways the media frames their coverage.
In our initial analysis, we take two similar and prominent cases of terrorism (Omar Mateen's shooting up the Pulse nightclub in Orlando and Dylan Roof's shooting up the AME Church in Charleston) and perform a variety of tests on them.
First, we gathered all the U.S. media articles we could on each of these attacks. Then we did an analysis of the most common language in the coverage for each, to provide us some broad context and possibly see any immediate differences. Then we ran STMs on each set of articles to see what kinds of topics arose. Finally, we ran an STM on the combined articles and performed some hypothesis testing to see if different topics were more closely associated with either Mateen or Roof.
Finally, to reflect more current events, we did a simple analysis of the content of the Charlottesville and Barcelona attacks from August 2017. These attacks were very similar, both involving cars driving through crowds with the intent to harm civilians. We pulled every article we could find and searched them for words associated with terrorism (informed by the results of our STM with Roof and Mateen).
Results
Our preliminary analysis suggests that there is differential media coverage. First, we found that while many of the words most cited in the Roof or Mateen articles were objective details of the attacks, the coverage pertaining to Mateen included several distinctly different keywords related to terrorism and religion.
Second, when running STMs on coverage of Mateen and Roof separately, we found more terrorism frames for Mateen than Roof (5 vs. 2). We also found the content of these terrorism frames to differ. Whereas Mateen's are more of what we think of as terrorism (common words refer to violence, ideology, etc.), the frames for Roof appear to be discussing whether his attack was terrorism or a hate crime. Thus, it is not quite the same, and these topics likely do not have the negative framing effect of Mateen's terrorism topics. This is also in spite of the fact that Mateen's ties to terrorist groups were very weak from the beginning, in contrast to Roof's clear ties to right-wing militant groups.
Third, when running a STM on the combined Roof and Mateen articles, we found several terrorism topics, as well as one that had explanations for the attack and one focused on mental health. We tested whether these frames were more closely associated with either attacker. If there was bias, we would expect the terrorism frames to be more commonly associated with coverage on Mateen, and the mental health and attack explanations frames to be more closely associated with coverage of Roof. This is precisely what we found.
Finally, for Charlottesville and Barcelona, we found articles about the Barcelona attacker were much more likely to use words like terrorism, terrorist, or religion than articles about the Charlottesville attacker. We plan to do much more with this, including running STMs, but given how recent the attacks were, the sample size is quite small for good statistical inference.
Figures and Tables
Word cloud of Dylan Roof Coverage in U.S. Media
Word cloud of Omar Mateen Coverage in U.S. Media
Google Sites
Report abuse