From Conversations to Communities: Aggregation of Latent Social Structures

By: Tina Cormier

Recent advances in computing, coupled with unprecedented data availability, have opened the door to a large new cohort of scientific questions waiting to be answered. Within the last decade, emerging methods in big data analytics have enabled social scientists to approach their existing research from novel perspectives as well as to branch off into new fields altogether.

Peter McMahan, a postdoctoral fellow at Stanford University and former University of Chicago Ph.D. student under James Evans, is an early adopter of this new technology. His research, which would not have been computationally feasible even five years ago, relies on text analysis of huge data sets in order to model and explain latent social structures.

McMahan has been associated with Chicago for over 10 years, first arriving at the Harris School of Public Policy for a Masters degree in 2005, before spending 2 more years as a research consultant and eventually earning his PhD from the Department of Sociology in 2017. He recalls the earliest days of his PhD, when there were very few resources for the kind of computational work he ultimately ended up pursuing.

“When I arrived, there were a handful of individual faculty doing this kind of work, but really no coherent path for students to learn big data methods for sociology,” says McMahan. “It was exciting to be part of such a dramatic transition that, today, finds UChicago as one of the big players in computational social science.”

McMahan’s dissertation research uses text analysis to dissect conversations in both face-to-face and online settings in order to demonstrate how subtle social signals within individual conversations, such as micro-perceptions of social status or power, are linked to macro-level structural inequalities.

At its core, the project relies on the well-established concept of mimicry, or accommodation, in interpersonal interactions. Humans tend to unconsciously and automatically adjust their posture, facial expressions, and even their style of speech to match behaviors observed in their communication partners. The way a person speaks – their linguistic style – is both measurable and closely attuned to the social context of the interaction.

In some settings, social mimicry is not balanced between the parties, such that one person alters their style significantly, while the other does not. Interpersonal communication is sensitive to contrasts in social status, meaning that people strive to affiliate with higher status individuals by adjusting their communication patterns to match those with higher status and minimizing mimicry of lower status individuals.

These patterns have been established for very formal power differences, as in those between supreme court justices and attorneys, but McMahan uses models of language structure based on millions of individual remarks to demonstrate that such power hierarchies are present on a smaller scale in just about any social interaction, and they both reflect and reinforce implicit structural inequalities in the greater community.

“Societal inequalities don’t only exist structurally or only for people who are overtly racist or sexist, for example, but they work their way into a lot of our personal interactions in a deeply ingrained way,” explains McMahan. A prevailing theme in his research is trying to reveal the understanding that individuals have about social structures that are not explicit or even conscious, but that are nevertheless apparent in the ways that people interact with one another.

His postdoctoral research with Dan McFarland at Stanford University follows a similar thread, but investigates a different kind of conversation – the kind that is played out in scientific journals. He is specifically interested in is how individual scientific fields are formed.

“It is remarkably slippery to define where a field came from, how is it changing, who has the power to change it, and who is more marginal,” he says. He is approaching the question with a relational framework of how authors relate to one another through citations. Specifically, he considers review articles as “field-forming” types of publications.

His findings reveal that review articles perform the important role of creating a legible translation of the field, though being cited by a review article frequently hurts an author, since future papers typically cite the review rather than the original publications. A smaller number of papers get a big boost as a result of a review article, and end up becoming central in the field.

“Review articles mark that moment where a collection of research becomes a subfield through this dramatic restructuring of the historical discourse in the literature,” says McMahan. The downside is that the authors who write the review articles have disproportionate power in shaping the conversation and the way the field progresses. “When you look at these journal articles as a conversation, it becomes possible to trace sets of cultural norms and assumptions among groups by looking at their interactions and how they relate to one another.”

McMahan is finishing his fellowship at Stanford within the next few months, and in July, he and his family will be moving to Montreal, where he will join McGill University as an assistant professor of sociology. He will be teaching a mix of introductory sociology and graduate level courses and is looking forward to teaching specialty courses on text analysis and relational methods.