Published: December 9^th, 2017

The study of radicalization continues to be dominated by theoretical and qualitative studies, with arguments regarding definitions and typologies commanding more space than actual evidence. The same is certainly true for the growing body of research pertaining to online radicalization. In much of the literature, the only numbers are the page numbers. Quantitative investigation appears to have been left to governments and security services with seemingly unlimited access to data. However, the results of government-directed research should also be questioned. The many tools and automated systems governments employ to identify online radicalization are not as efficient as they may claim to be. In this respect, automated identification tools often result in false arrests, on the one hand, while simultaneously failing to identify real and immediate threats, on the other. It has been argued that one of the reasons that such systems are not as efficient or successful as they ought to be is because they are not evidence based. That is, an algorithm is only as accurate as its input, and the inputs on which many systems have been designed have often been based on purely theoretical assumptions. 

The challenge for any security service and for tool designs is how to identify the small percentage of imminent threats within a much larger pool of highly radicalized but non-violent individuals. It is an axiom of science that it is impossible to assess sample bias from looking at the sample alone. So too, it is impossible to identify characteristics that are unique to terrorists solely by examining terrorists or comparing them to each other. In this regard, Joshua Frielich, Gary LaFree, and other leading scholars in the quantitative study of radicalization and terrorism emphasize the need to compare terrorists with non-violent radicals of the same persuasion. Again, since only a small percentage of those holding radical beliefs will go on to commit violence, it is important to identify what characteristics differentiate between these two outcomes of radicalization, rather than what distinguishes them from the general population. 

In our research we have been coding and analyzing the social media behaviors of dozens of terrorists in the 100-day lead up to their attacks. Each terrorist has been matched with a non-violent radical from the same network in order to create the comparison group. So far we have been examining differences in types of activities and other social media level metrics derived from social learning theory. While we have already made many interesting and important findings, few are as interesting as what we found regarding the types of posts being made by radicals.

In the current study we coded for 10 different types of activities, such as text post, image post, video post, etc., and text share, image share, video share, etc. One of the most striking differences is that the non-violent radicals have a significantly greater proportion of original written text posts. This means that non-violent radicals are more likely to post lengthy expositions of their radical beliefs. They articulate themselves, using their own words, almost twice as much as those who go on to carry out terrorism do. The magnitude of the differences is increased by the fact that the non-violent radicals also have fewer posts per day overall compared to the terrorists. Conversely, the terrorist group displayed a significantly higher proportion of “sharing” of radical images and videos, as opposed to original uploads. In this regard, when sharing images and videos, the non-violent radicals are also more likely to add their own thoughts and commentary.

The research team noted that these findings may support theories suggesting that social media can provide an outlet that reduces the risk of violence. This theory posits that by providing a non-violent venue for expressing grievances, and as a platform in which individuals can feel they are contributing from the comfort of their own home, radical social media usage may reduce the likelihood of violence for some percentage of the at-risk population. In this respect, the initial findings from our research seems to suggest that when it comes to online radicalization, those who bark the loudest are the least likely to bite.

This does not mean that those who articulate themselves and use social media as an outlet to voice radical ideas never turn to violence; some certainly do. Rather, the identification of such outcome-specific online behaviors provides additional parameters facilitating the identification of violent individuals.