UMD Researchers Follow the Money to the House Floor

While previous research has struggled to document a connection between money and congressional votes, a new University of Maryland study shows a link between donations and what members talk about in floor speeches.

The multidisciplinary team of researchers from several institutions in the United States and Canada used powerful computational tools to analyze texts of floor speeches, records of donations from political action committees and other data from 1995 to 2018. Its findings map out robust connections between political speeches and donor behavior in the U.S. political system and were published yesterday in the journal PLOS ONE.

The new study zeroing in on the topics and language of the speeches provides more room for nuanced study than surveying just the binary yes-no choice in voting, said lead author Pranav Goel M.S. ‘20, Ph.D. ’23.

“In fact, donor behavior offers a significantly better explanation for politicians’ policy focus when they talk, compared with other relevant factors, such as politicians’ home state or even their party affiliation,” Goel said.

To conduct the study, the team developed a machine learning framework that used metadata on every member of the U.S. House of Representatives in the 23-year study period—a voluminous set of data that in addition to the speeches and donation records also included committee assignments and background information about PACs, such as what industry they are affiliated with.

Data on floor speeches was pulled from the Congressional Record, while raw data on U.S. political campaign contributions came from public sources including the Federal Election Commission and OpenSecrets, a nonprofit research organization that tracks the flow of money in U.S. politics. Through a combination of automated and manual efforts, the two databases were carefully linked by the research team, and computing resources in the University of Maryland Institute for Advanced Computer Studies helped build, manage and crunch this data in order to enable this research.

Because the machine learning tools that the team used can process such massive amounts of data, more information is now available to political scientists on this topic than ever before, said co-author Kristina Miler, associate professor of government and politics at UMD.

“We’ve always had the sense that money is ‘doing something’ when it comes to shaping the issues being discussed and how they’re being discussed, but there’s been no hard evidence of that,” Miler said. “Our team wanted to go beyond conventional wisdom. The interdisciplinary collaboration between computer scientists, linguists and political scientists on this project is unique and very exciting.”

This research was supported by funding from the National Science Foundation’s Division of Intelligent Information Systems.

“What's truly exciting about this research is that it combines advances in technology with progress on important, substantive questions about the way American democracy works," said co-author Philip Resnik, MPower Professor in Linguistics and the Institute for Advanced Computer Studies. "The study uniquely combines rich and careful data collection—the lifeblood of data-driven research—with a rigorously formulated and validated computational model for issue attention. Especially at this point in time, I'd say new insights into the relationship between money and our legislative process are about as high-impact as you can get.”

The research team emphasized that, when taken together, its framework, data and findings can help not only to answer research questions political scientists care about, but to increase the transparency of the role of money in politics.