Social Science with Social Media

Abstract

Over the past twenty-five years, electronic communication has matured from being a niche social activity mainly enjoyed by academics and engineers, to an important enabler of the daily activities of a demographically diverse population of hundreds of millions of people worldwide, augmenting, complementing, or even replacing offline methods of socializing, dating, shopping, learning, working, and engaging in political activities. While this transition is in itself interesting, it also has significant implications for the social and behavioral sciences. The daily lives of people worldwide are now captured in detailed, event-level recordings that are time-stamped and geo-stamped, providing researchers with a new kind of observational data, enabling them to address fundamental questions about social identity, status, conflict, cooperation, collective action, and diffusion. This dissertation explores these implications, with a critical review and two empirical explorations. Chapter one reviews existing literature and examines the methodological challenges that arise along with the opportunity provided by online behavioral data, including generalizing to the offline world, protecting privacy, and solving the logistical challenges posed by data at a larger scale than social and behavioral scientists typically use. Chapter two is an investigation into measuring the rhythms people experience in their mood over the course of the day, week and year. By analyzing the text of hundreds of millions of timestamped messages from the social media service Twitter, I show that there is a consistent shape to people’s moods over time, including boosts in positivity in the morning and on the weekend, and that seasonal variation tracks changes in daylength. Finally, chapter three examines how moral judgments about personal debt affect decisions by lenders about who to lend to and at what rate. By analyzing the text portion of loan applications in the microlending service Prosper.com, I show that, though traditional economic characteristics like credit score dominate, non-economic characteristics also help predict lending outcomes and have effects that are mediated by the creditworthiness of the applicant.