Please register for the CV & Online Identity Workshop below.

The emergence of social media has changed many aspects of society, but one of the most stark changes has been the transformation of the social sciences from a data-poor field of research into one flooded with data. Revolutions in computation, digital storage, and connectivity have produced large, centralized warehouses of data on human behavior. With the advent of online social media, it became possible to store previously unimaginable amounts of behavioral data on individuals and make this data accessible to a wide array of social scientists. However, the sudden flood of data has not always led to the construction of meaningful measures of social behavior. This dissertation focuses on techniques for effectively using large-scale digitial trace data to study American political behavior.

I begin by discussing how record linkage of social media and administrative data can support large-scale analyses of online behavior at scale, with modest bias (Chapter 1). I demonstrate the utility of this approach by applying the dataset to study the effect of Twitter's deplatforming of right-wing extremist accounts, one of the first such studies of the causal effects of platform-level interventions on user behavior (Chapter 2). I next turn to concerns regarding the measurement of news sharing, arguing that conflicting results in the literature on "filter bubbles" is due to a lack of clarity on when domain-level analyses are preferable to URL-level ones (Chapter 3). Turning to the misinformation literature, I use behavioral data from Facebook to argue that many studies of fake-news sharing rely on flawed operationalizations of fake news or disinformation, failing to account for the role of satirical websites (Chapter 4).