Dahyun Choi

Hello! I am a Ph.D. candidate in the Department of Politics at Princeton University, where I am affiliated with the Data-Driven Social Science Initiative and the Program for Quantitative and Analytical Political Science. I work on questions in interest group politics, lobbying, regulation, climate change, and American political institutions, using methods from computational social science and applied statistics. I also develop methodological tools to improve machine learning practices in political science.

Working Papers

Teaming Up Across Political Divides: Evidence from Climate Regulations

(Previously titled "Teaming up with the Enemy: Firms and the Information Environment of Climate Regulations."). Under Review.

[ | paper | ]

Why do interest groups with contrasting interests and policy goals work together? This paper argues that interest groups prioritize high-quality implementation of policies even when it means compromising on their policy preferences. To test this argument, I introduce original measurement strategies to reveal the systematic patterns in which firms and environmental groups invest in joint efforts to improve the implementation of greenhouse gas emissions standards. The analysis, using public comments spanning 2010-2020, demonstrates that comments written by joint efforts of environmental groups and firms contain more informative text that can contribute to a higher quality of policy implementation, compared to individual efforts alone. While the compromise is biased in the direction of firms' interests, environmental groups can exercise meaningful influence over the finalized policy outcome by inducing more participation from the former. These findings highlight the hidden dynamics of regulatory politics, wherein divergent political goals are reconciled for high-quality policy implementation.

Frontiers in Money in Politics Research, University of Southern California

2023 American Political Science Association, Los Angeles

Money in Politics Conference, Copenhagen Business School

Climate Pipeline Project, Brown University

2023 Midwest Political Science Conference, Chicago

2022 Environmental Politics and Governance, Virtual conference

2022 International Political Economy Society, University of Pittsburgh

Quantifying Informative Signals from Political Texts

(with Brandon Stewart and Denis Peskov). Under Review.

[ | | ]

How do interest groups use political communication to appeal to the general public? Classical accounts of interest group politics have assumed that interest groups increase voters’ knowledge about policy by informing them of expected consequences or the contents of policy. However, interest groups tend to mix two kinds of signals that are difficult to separate empirically: 1) informative signals that help voters improve political decisions, and 2) persuasive signals that influence the formation of voter preferences but don’t have direct relevance to the substantive question of interest. We create an eight-category typology for these signals and generate high-quality manual annotations of public statements and press releases distributed by interest groups. We develop an automated coding procedure using recently developed large language models, which can classify the presence of signals beyond the capacity of non-expert human coders. Our findings, drawn from illustrative examples of congressional hearings and public advertising, suggest that interest groups strategically tailor the composition of signals under different institutional settings.

Email for current draft

2023 American Political Science Association, Los Angeles

2022 Society for Political Methodology, Washington University in St. Louis

2023 International Studies Association, Montréal

Learning from Noise: Applying Sample Complexity for Political Science Research

(with Perry Carter)

[ | paper | ]

Statistical learning increasingly provides a bridge between theoretical concepts and empirical realities in quantitative political science research. However, many concepts of interest to scholars are both highly multidimensional and inherently fuzzy, making manual classification highly error-prone. The consequences of these issues for machine learning applications are not well understood, generating persistent uncertainty over what constitutes good enough data. In this paper, we introduce a novel framework, learning from noise, taking advantage of researcher-specified bounds on conceptual complexity and labeling error to guarantee the sample size necessary to achieve a minimum level of accuracy with a precise level of confidence. We provide the scR R package online, offering a computationally efficient way to implement the proposed methods. While measurement built on clean data with well-defined concepts will always be encouraged, we aim to improve standard practices by providing a general-purpose tool to assess the feasibility of applying statistical learning models.

2023 American Political Science Association, Los Angeles

2023 Society for Political Methodology, Stanford University

Scientification of Politics? Interest Group Influence on Bureaucratic Expertise in Climate Change

[ | | ]

Science is considered the primary source of information that helps regulators improve policy outcomes. However, the use of scientific information is inevitably intertwined with political reasoning. In this paper, I construct a novel dataset revealing the internal practices of science evaluation by the Environmental Protection Agency and examine how suggestions from interest groups influence the way bureaucrats evaluate science. My empirical evidence shows that scientific studies recommended by industry are more likely to be included in final documents, regardless of subject-normalized scientific impact indicators. Using unsupervised methods, I further show that industry-endorsed scientific studies are more likely to be referenced to justify describing air pollution exposure as having no effect on public health. This paper not only provides suggestive evidence on the internal decision-making of bureaucrats, but also contributes to our understanding of the political influence of interest groups under scientific uncertainty.

Email for current draft

Princeton Institutions Working Group

2023 International Political Economy Society, Georgetown University

Investing in Team Efforts: When Agents with Contrasting Preferences Collaborate

[ ]

I develop a model of policymaking in which policy experts with asymmetric capabilities invest in joint efforts to craft high-quality proposals. While experts may have different policy preferences, they share a common interest: achieving high-quality policy implementation. Agents have asymmetric capabilities and their actions generate outcomes that vary across two dimensions: i) a distributional dimension that reflects how the divergent policy goals of the agents are reconciled, and 2) a quality dimension that represents a set of features that enhance the efficacy or equity of policy implementation. The model demonstrates that winning the ideal policy is purely instrumental as long as experts can achieve a higher quality of policy implementation, which makes compromise arise endogenously.

How Interest Groups Strategically Take Advantage of Voter's Rational Inattention

[ ]

2022 Midwest Political Science Conference, Chicago

2020 Society for Political Methodology, Virtual conference

A Dataset of Presidential Election Results at the Congressional District Level

(with Rocio Titiunik, Francesca Tang and Christian Baehr)

Teaching

Preceptor, POL 574 Quantitative Method IV

[ ]

This is the most advanced graduate-level course in the politics method training sequence. The class covers a variety of supervised and unsupervised learning techniques, including both mechanical and probabilistic models, and computational methods for the analysis of new types of data, including text, networks, audio, and images (instructor: John Londregan).

Preceptor, POL 345 Quantitative Social Science

[ ]

This is the undergraduate-level politics method training sequence. The class covers a variety of topics in statistics and causal inference (instructor: Marc Ratkovic).

Preceptor, POL 245 Visualizing Data

[ ]

This is the first course in R and basic statistics for incoming freshmen in Princeton. It is one of the classes offered for Freshman Scholars Institute (instructor: Brandon Stewart).