I study presidential power and policy choice under institutional constraints. My research examines which policies presidents pursue through unilateral action and how far those actions move policy from the status quo. My dissertation develops methods to classify presidential directives and measure their ideological content.
My methodological work focuses on measurement and validation in computational social science. I study how evaluation design shapes the choice of text classifiers and how faithfully survey digital twins reproduce human response patterns.
I am affiliated with the Laboratory for Technology, Communication and Democracy (TeCD-Lab) at UC Riverside, directed by Professor Kevin Esterling. I also serve on the web development team for Prytaneum, a platform for deliberative public engagement.
This dissertation examines how presidents calibrate the content of unilateral actions to manage institutional constraints and maximize political benefits. The first chapter distinguishes ideological from non-ideological directives, then develops anchored embedding scaling (AES) to estimate their locations within liberal-conservative policy space. The second chapter recovers the inherited policy status quo for each directive and estimates the policy displacement produced by unilateral actions using AES. The third chapter extends existing models of unilateral action by proposing a cost-constrained model of presidential policy choice.
Chapter 1
Classification and AES
Part I develops a classification of presidential directives that distinguishes ideological from non-ideological actions. Part II uses anchored embedding scaling to locate directives in liberal-conservative policy space, combining text embeddings with ideological estimates for legislative bills.
Chapter 2
Status Quo and Policy Displacement
This chapter recovers the policy status quo from which presidents depart when they act unilaterally. It develops a pipeline for identifying the relevant status quo against which presidential directives can be interpreted as policy movement. Using anchored embedding scaling, the chapter estimates the policy displacement produced by unilateral actions.
Chapter 3
Theory and Institutional Constraint
This chapter develops a cost-constrained model of unilateral action. It tests the theoretical prediction that presidents are likely to move policy toward the veto pivot, while arguing that presidents may stop short of this constraint boundary when institutional costs become sufficiently high. The chapter examines why presidents limit the policy displacement of unilateral actions even when unilateral authority allows them to move policy on their own.
My research brings together political institutions, political communication, and computational methods to study political actors' strategy and representation.
with Eunseong Oh
Invited to resubmit, Political Analysis
We compare five classifiers across three congressional text tasks to examine how test design shapes model choice. The results show why researchers should define the population they intend to classify before selecting a model.
Text Analysis · NLP · Model Selection
Dissertation Project · Chapter 1
I develop a classification of presidential directives that distinguishes ideological from non-ideological actions. This distinction connects the measurement of unilateral action more closely to the policy choices that theories of presidential power seek to explain.
Presidential Power · Unilateral Action · Classification
Dissertation Project · Chapter 2
I develop anchored embedding scaling to locate presidential directives in ideological space. The method combines text embeddings with ideological estimates for legislative bills to place unilateral actions on a common scale.
Ideological Scaling · Text Embeddings · Measurement
with Eunseong Oh
Fidelity Validation of Survey Digital Twins
We evaluate how faithfully LLM-based survey digital twins reproduce human response distributions. Using ANES and CES data, we examine whether persona attributes selected for one survey question remain useful across questions, substantive domains, and survey instruments.
Survey Research · Digital Twins · Validation
Under review, American Politics Research
Examines whether conspiracy beliefs shape campaign contributions in American politics, using the 2012 and 2016 ANES. The paper links conspiratorial belief to donor behavior across partisan contexts.
Campaign Finance · Conspiracy Theories · ANES
with Eunseong Oh
Presented at PolMeth 2025
Uses CLIP technology to analyze how political videos portray megadonors. Applies multimodal deep learning to understand the framing of political visual media.
Multimodal Analysis · CLIP · Political Communication
Web Development Team
Affiliated with TeCD-Lab, UC Riverside
I contribute to the development of Prytaneum, an AI-assisted platform for deliberative public engagement. Developed at the Laboratory for Technology, Communication and Democracy, directed by Professor Kevin Esterling, the platform helps speakers understand participants' ideas and perspectives at scale.
Coming soon
Instructor
UC Riverside
Instruction in foundational statistics and R for incoming graduate students.
Guest Lecturer
UC Riverside
Guest lectures on data literacy, research presentations, poster design, and abstract writing for the MacREU/MEMENCYS summer program.
UC Riverside (2023 – Present)
Vocabulary separability diagnostics for text classification.
Companion package for Oh and Yu, “What Are We Generalizing To? How Test Design Shapes Model Choice in Congressional Text Classification.”
Explore 24,625 presidential directives on a liberal-conservative scale, with ideology filters, party distributions, presidential means, and NOMINATE validation.
Try it →Vocabulary separability diagnostic for text classification. Measure Jaccard overlap between classes to decide: sparse (TF-IDF) or dense (neural)?
Try it →Explore 44,345 amendments to Korean presidential decrees (대통령령) across nine administrations, from 1988 to March 2026. Filter by president, ministry, and amendment type.
Try it →How different is the language Democrats and Republicans use on the floor? Eight decades of vocabulary divergence across the 79th-119th Congress (1945-2026), with chamber splits, term trajectories, and era comparison.
Try it →