Research

I am both a scholar of American political behavior and an applied methodologist specializing in computational social science and causal inference. Three core goals bridge these two fields into a cohesive research agenda that uses novel applications of computational and empirical methods to better understand how the American public perceives and reacts to ideological policymaking during a divisive era in U.S. politics.  I also apply these insights to better understand how humans make sense of politics, particularly on contentious ideological topics. These complementary pillars of my methodological research supplement my substantive research agenda on American political behavior, which focuses on the themes of ideological polarization and extremism, state politics and policy, political geography, and democratic health in the United States today.

Pillar 1: How does combining LLMs with non-AI machine learning methods can help us understand how LLMs make decisions? How can this help advance public opinion research in political science?

A major focus of my current research on LLM decision-making investigates human-AI alignment in experimental choice contexts. My solo-authored work on this subject (forthcoming at Public Opinion Quarterly) is, to my knowledge, the first peer-reviewed study in political science to systemically evaluate how LLM-based silicon samples (Argyle et al. 2023) simulate live human subjects in forced-choice conjoint tasks. This analysis establishes a 'ground truth' of model performance in conjoint experiments by testing whether pre-trained, 'out of the box' LLMs benchmarked to large samples of live respondents can simulate diverse human preferences on hard-hitting ideological issues in American politics.

The results of this project directly inform my current and future work at the intersection of human-AI alignment and American public opinion. I argue that we can use interpretive machine learning methods to glimpse inside the black box of LLM reasoning to pinpoint where human-AI alignment breaks down. This work, which uses tree-based interprative machine learning strategies as a validation metric for silicon samples, has been accepted as a book chapter in the forthcoming edited volume Doing Valid Social Science with Large Language Models. I am currently generalizability testing these new strategies for glimpsing inside the black box of human-AI alignment to understand more about how LLM and human decisionmaking processes differ from each other, and what these differences imply about identity-based reasoning. 

Pillar 2: How can swe use LLMs as tools to better understand how humans make sense of political issues and processes within American politics?

I am in the early stages of an interdisciplinary collaboration that uses LLM agents to simulate democratic discourse in New England town hall-style direct democracy settings. This project will offer insight into the potential promises and pitfalls of using AI agents to enhance political participation in direct democracy contexts and present openings for future work on how scholars and citizens alike might use LLMs to promote constructive democratic deliberation and enhance governmental responsiveness. I am also working on several other projects that speak to these themes, including using AI agents to model American voters with different identities and varying degrees of ideological sophistication. 

Substantive Research: How do Americans make sense of politics and policy today, particularly on contentious ideological topics?

The objectives above complement my robust substantive research agenda, which focuses on American public opinion and political behavior. My work especially centers the themes of ideological polarization and extremism, state politics and policy, political geography, and democratic health in the United States today.

Over the last decade, state legislatures have passed an increasing number of policies on hard-hitting social and cultural issues, including abortion, gun control, and LGBTQ+ rights; because these laws regulate deeply personal and contentious issues, they hold distinct implications for democratic responsiveness. These stakes underscore the importance of understanding how Americans make sense of contentious policies in today's era of ideologically driven policymaking. In a solo-authored paper forthcoming at State Politics & Policy Quarterly, I argue that these hard-hitting state policies raise the stakes on where individuals choose to live because they explicitly dictate what residents can and cannot do, using a preregistered conjoint experiment of 29,000 respondents paired with an original dataset composed of 30,000 recent movers and state-level abortion policies, I find strong experimental and modest observational evidence that such laws may drive interstate migration in the post-Roe v. Wade era. 

Other projects explore public opinion on controversial ideological issues. Several papers in progress leverage novel perceptual data on state-level abortion laws to untangle how the 2022 Dobbs v. Jackson Women's Health Organization Supreme Court decision, that overturned Roe. v. Wade, impacted democratic responsiveness in state governments.

I also use sophisticated causal interence designs to answer long-standing questions related to political participation and state-level democratic institutions. A solo-authored article published in Election Law Journal tests whether the Supreme Court’s 2013 Shelby County ruling led to electoral backsliding in once-covered states, finding no evidence that backsliding occurred from the repeal of these preclearance regulations. Another manuscript, currently under review at Nature, evaluates how the March for Our Lives gun control advocacy movement mobilized young voters under 30 in the 2018 midterms, finding evidence that this campaign led to higher registration and turnout among general election voters. Importantly, these results imply that peer-to-peer mobilization efforts may boost youth participation. These findings suggest that peer-led advocacy campaigns on salient issues can help close longstanding generational gaps in political participation.

Finally, I specializing in the use of large-N survey experiments, causal inference identification strategies, and large datasets to answer important substantive questions in political science and improve survey methods in public opinion research. A new book chapter in the forthcoming edited volume Understanding Experimental Political Science: Recruiting and Engaging Diverse Participants discusses the effective analysis of rare subgroups in survey experiments using tree models to detect attribute-level heterogeneity in conjoint experiments. I find that this method maintains the integrity and robustness of treatment effect estimates for rare trait groups without the power losses traditionally associated with small subgroup analyses. 

You can find a full overview of my research here and links to publications and working papers here.