Computational Social Science

Experimentally-based social science today is moving slower and finding less than it could be. Although recent advances in digital technologies and crowdsourcing services allow individual experiments to be deployed and run faster than in traditional physical labs, a majority of experiments still focus on one-off results that do not generalize easily to real-world contexts or even to other variations of the same experiment. To achieve replicable, generalizable, scalable and ultimately useful social science a fundamental rethinking of the model of virtual-laboratory style experiments is required. Not only is it possible to design and run experiments that are radically different in scale and scope than was possible in an era of physical labs; this ability allows us to ask fundamentally different types of questions than have been asked historically of lab studies. However, taking full advantage of this new and exciting potential will require four major changes to the infrastructure, methodology, and culture of experimental science: (1) significant investments in software design and participant recruitment; (2) innovations in experimental design and analysis of experimental data, (3) adoption of new models of collaboration, and (4) a new understanding of the relationship between theory and experiment. The Virtual Lab project pursues this ambitious path to facilitate a new class of scientific advances in our understanding of social phenomena.

Since the 2016 US presidential election, the deliberate spread of misinformation online, and on social media in particular, has generated extraordinary concern, in large part because of its potential effects on public opinion, political polarization, and ultimately democratic decision making. However, proper understanding of misinformation and its effects requires a much broader view of the problem, encompassing biased and misleading–but not necessarily factually incorrect–information that is routinely produced or amplified by mainstream news organizations. Project Ratio measures the origins, nature, and prevalence of misinformation, broadly construed, as well as its impact on democracy.

Speech and dialogue are the heart of politics: Nearly every political institution in the world involves verbal communication. Yet vast literatures on political communication focus almost exclusively on what words were spoken, entirely ignoring how they were delivered—auditory cues that convey emotion, signal positions, and establish reputation. We develop a model that opens this untapped information to principled statistical inquiry: the model of audio and speech structure (MASS). Our approach models political speech as a stochastic process shaped by fixed and time-varying covariates, including the history of the conversation itself. In an application to Supreme Court oral arguments, we demonstrate how vocal delivery signals crucial information— skepticism of legal arguments—that is indecipherable to text models. Results show that justices do not use questioning to strategically manipulate their peers, but rather engage in genuine fact-finding efforts. Our easy-to-use R package implements the model and many more tools for audio analysis.

Why have body-worn cameras failed to live up to their hype as transformative tools for police accountability? In conversations with police commanders, we consistently hear that agencies lack the capacity to manually review hundreds or even thousands of hours of footage, nearly all of which is deleted before review due to the financial strain of video storage. We are working to develop automated techniques for analyzing massive-scale footage from police body-worn cameras (BWCs). With university seed funding, our team of computer-vision, speech analysis, and social science experts is annotating videos to train artificial intelligence algorithms, evaluating state-of-the-art analytic tools, and communicating closely with police departments to develop practical tools for police supervisors.

Based largely on vivid anecdotes, a media narrative has taken hold ascribing worrisome power to hidden algorithms. Academic researchers have struggled to add evidence to this debate due to both data limitations and serious research design challenges. We undertake a novel experimental approach that seeks to resolve the impasse on this critical question by randomizing the recommendations themselves. Our team has built an interface that serves sequences of videos using the recommendation tree from YouTube’s API. We create our own variations on the existing algorithm by injecting differing proportions of cross-cutting video content (algorithmically matched according to several criteria) to randomly assigned groups of subjects. This approach enables an attractive combination of experimental control and external validity without the need for reverse-engineering YouTube’s blackbox recommendations. Subjects in our study will choose to watch a sequence of short videos and follow recommendations while the interface tracks watch time and other measures of engagement. After the viewing session, participants will be surveyed on issue opinions, affective polarization, and perceptions of media hostility to their preferred viewpoint. If online video recommendation algorithms drive affective polarization, our design will provide rigorous evidence of the extent and magnitude of the effect.