Skip to main content

Feature Stories

Networked Decision-Making
Analyzing big data to understand complex social systems
Network viisualization of political action committee contributions to candidates

Desmarais' research methodologies are being used to shed light on collective decision-making processes, from the use of international sanctions to the use of scientific expertise in policymaking.

In our increasingly digital age, the collection, storage and analysis of large-scale data offer a window to the inner workings of our social world and collective interactions. At UMass Amherst, computational social science expert Bruce Desmarais (political science) is bridging these worlds by developing quantitative methodologies to help explain the complex ways in which governments and organizations make collective decisions.

Desmarais’s research in network analysis has shed light on a wide range of issues. From voting patterns in the US Supreme Court, to understanding legislative collaboration, to analyzing international conflicts and government information management, his work reveals the intricacies behind public decision-making.

Along with political science colleague Raymond La Raja and graduate student Michael Kowal, Desmarais is investigating the hot-button issue of campaign finance in the House of Representatives. Specifically, they are examining the role of political party networks in supporting candidates and shaping electoral outcomes. In evaluating how support from interest groups affects challengers’ chances in elections, Desmarais and his colleagues have found the use of networks improves their challengers’ electoral prospects.

Bruce Desmarais, Political Science
Desmarais is also working with public policy scholar John Hird to better understand how science is used in policymaking. To accomplish this, they are studying a collection of impact analyses produced by regulatory agencies to justify cost and explain the benefits of various US regulations. Their research shows that some agencies employ science extensively in their analyses, yet use varies greatly between agencies.

On an international level, Desmarais is working with University of North Carolina political scientist Skyler Cranmer and University of South Carolina political scientist Tobias Heinrich to study what they refer to as the “international sanctions network.” He explains how methods used by countries and international organizations to strategically coerce each other with economic hardship can make for hostile relationships. In gathering data regarding international tendencies to sanction and be sanctioned, Desmarais and the team pay particular attention to how countries respond to being sanctioned. They have found that sanctioned countries often retaliate within one year of being targeted, and if not during that timeframe, they most often do not retaliate at all.

Desmarais is also working with Cranmer to develop models that can reasonably predict the flow of transnational terrorism. Using a database with over 12,000 transnational terrorist attacks, they have tracked terrorist patterns using statistical models that forecast both countries of origin and target countries for impending attacks—information that could be used to more efficiently allocate security resources.

In other work Desmarais is collaborating with computer scientist Hanna Wallach to apply machine-learning algorithms and social network analysis to archived government materials. Their aim is to examine how issues move along the continuum from informal lines of communication, through public meetings and into legislation. Drawing on public records—archived government emails, public meeting agendas and minutes, and legislative documents—Desmarais explains that the methodology by which they collect the data is a critical element of the research.

Once the archived emails have been collected, Desmarais and Wallach apply a series of machine learning algorithms that enable them to “parse the emails by topic.” These topics are not pre-specified by the researchers, but rather the algorithms tell the machine what to do. The computer is then programmed to find repeated vocabulary and group emails together using key words. One use of this method is to study how intra and inter gender communication patterns vary with the topics of discussion.

“Previous findings in this area indicate that women are on the periphery,” Desmarais says. “They’re not as central in professional communication networks, and what we’re adding to this literature is looking at whether that varies by topic.”

In a related project, Desmarais and Wallach are looking at the internal communications that lead to the declassification of documents and ways the process could be made more efficient. Millions of documents, such as those related to collective bargaining and executive meetings, must go through the declassification process on an annual basis—a time consuming process that involves a great deal of manpower. Desmarais is working on statistical models that could help government officials declassify documents based on topic. This would free up government resources and allow officials to release information to the public more quickly.

Desmarais is a core faculty researcher in the campus’s Computational Social Science Initiative, a collaborative organization of scholars in computer science, political science, sociology, and statistics. The initiative merges big data analytics with the social sciences to facilitate the investigation and understanding of complex societal systems.

“It’s fairly complex in that it’s not simple, numeric data that fits into a nice spreadsheet,” Desmarais says. “It’s messy, it has text and numbers and social, individual identities.” 

Amanda Drane '12

Banner image (courtesy of Bruce Desmarais)a network visualization of political action committee contributions to candidates in elections for the US House of Representatives