home | cv | dissertation | papers | resources
In non-democracies, it is widely recognized that the decision to participate in anti-regime collective action carries high costs. Individuals possess limited information with respect to the population-level distribution of pro- and anti-regime sentiment. And yet, this problem is not insurmountable and high-risk collective action occurs, as recent events in the Middle East and North Africa emphasize. Granovetter, Kuran, Lohmann, and others have demonstrated that this informational deficit can be partially overcome by observing others' participation in protest. These theories of collective action take as their starting point a group of initial participators (usually committed extremists, although Lohmann's argument stresses activist moderates) whose action then convinces fence-sitters. Yet, as Tilly, Tarrow, Meyer, and McAdam have argued, protest requires political opportunities -- and potential collective actors must have information to perceive opportunities.
I consider the "Arab Spring" that emerged unevenly across the region in 2010-2011 and attempt to build upon these two traditions of collective action theory. Threshold models often take some participation as given and information revelation prior to the onset of protest as exogenous. A primary and overlooked source of this information is the media. In non-democracies, the media is discounted due to perceived censorship and state interference, yet in actuality few totalitarian societies remain that completely manage information flows. Instead, there often exists a range of media outlets with a distribution of pro- and anti-regime alignments. As individuals self-select into their media consumption, news that is consonant with these assumed alignments is uninformative with respect to the population distribution of regime preferences. However, deviations from expected content and tone are highly informative moments for both citizens and regime. I contend that these moments provide potential collective actors with information to update their beliefs about both the distribution of regime preferences and the regime's vulnerability. The extent to which the regime is able to manage or prevent these deviations is an important signal as to its capacity to repress. Note that this has the counter-intuitive implication that regimes with highly restrictive media environments are at increased risk of even minor deviations being highly informative--an argument that I develop in another chapter of the dissertation.
Identifying and measuring these moments presents a methodological challenge that I address using an original data source and quantitative text analysis methods imported from computer science. My data are daily full-text articles in both Arabic and English from major state-run and independent media outlets in Egypt and Jordan from the months and years (coverage varies by source) before, during, and after the so-called "Arab Spring." I use Latent Dirichlet Allocation (LDA) to model the evolving topical structure of the media; LDA is a generative probabilistic modeling algorithm that estimates the distribution of topics within a corpus of documents, the distribution of topics within a single document, and the distribution of words within a topic. In doing so, I condition on media source, language, and date to track the convergence and divergence of media attention to identify moments of increased information. I use two measures, developed by Hall, Jurafsky, and Manning (2008): topic entropy, or how diffuse or concentrated coverage is with respect to the distribution of topics, and the Jensen-Shannon distance, a similarity metric to compare probability distributions. Data collection and analysis are ongoing.
I am always seeking to collaborate with others on related topics, in both the social sciences as well as in computer science.
If you are interested in discussing such a collaboration, please e-mail me at tcausey [at] uw [dot] edu.