Publications

The complexity of legislative language is of theoretical importance to many substantive questions about legislative politics. However, most existing measures of bill complexity are either generated at the broad issue level and applied to individual bills, or they are reliant on a simple metric like length. In this paper, we apply a pairwise comparison framework to the measurement of complexity in legislative texts. We compare the results of a Bradley-Terry model fit on pairwise comparisons made by human coders with the results of the same model fit on comparisons made by large language models (LLMs). There is a moderately high level of agreement between human coders and the LLMs, and the relationships between observable text features and the underlying trait of complexity are similar in comparisons made by human coders and by the LLMs. Our work demonstrates that, with researcher-selected bridging texts and carefully designed prompts, LLMs can be used to measure complexity in legislative texts.

The theory and practice of what has come to be called "deliberative democracy" have been revived for the modern era with a focus on deliberative microcosms selected through random sampling or "sortition." But might it be possible to spread some of the benefits of deliberation beyond mini-publics to the broader society? Can technology assist with scaling an organized deliberative process? In particular, would those who experience such a process become more deliberative voters? Would their considered judgments from deliberation influence their voting? We draw on a larger than usual experiment with public deliberation and a one-year follow-up in the mid-term U.S. elections to suggest answers to these questions. It has implications for whether spreading an organized deliberative process could, in theory, be used to create more deliberative elections.

Delegation of powers from the legislature to the executive branch is a nearly ubiquitous feature of modern lawmaking. However, much of what scholars know about delegation is gleaned from an exclusive focus on landmark legislation. We introduce a method to identify delegating language across a larger universe of legislation. Using an active learning convolutional neural network on bill text, we classify bill sections by their delegation content, applying an iteratively improving coding scheme that enhances existing supervised learning approaches. We develop a novel dataset that allows us to answer important questions about interbranch relations. First, we find that legislator ideology, partisanship, and institutional position affect the delegatory content of introduced legislation. We then explore the role of delegation in the advancement of bills through the legislative process. Finally, we evaluate the ally principle, finding that variation in delegation is driven by cross-agency differences in ideology and structural independence.

Does deliberation produce any lasting effects? "America in One Room" was a national field experiment in which more than 500 randomly selected registered voters were brought from all over the country to deliberate on five major issues facing the country. A pre-post control group was also surveyed on the same questions after the weekend and about a year later. There were significant differences in voting intention and in actual voting behavior a year later among the deliberators compared to the control group. This article accounts for these differences by showing how deliberation stimulated a latent variable of political engagement. If deliberation has lasting effects on political engagement, then it provides a rationale for attempts to scale the deliberative process to much larger numbers. The article considers methods for doing so in the context of the broader debate about mini-publics, isolated spheres of deliberation situated within a largely non-deliberative society.

This article addresses the interstate differences in outcomes from the coronavirus disease COVID-19 pandemic by focusing on state capacity. State capacity refers to states' ability to create and implement policy. We posit that states want to limit death and destruction within their borders. COVID-19 created an instance in which states had a shared, preferred outcome but had very different levels of success. Using a novel measure of state capacity that allows for subnational comparisons - and is independent of ideological political will - we show that states with greater capacity experienced fewer excess deaths during 2020 and more successfully distributed vaccines in early 2021. The findings are robust to various measures of partisanship, social capital, geography, and demographics. Our work bridges US state politics literature and comparative politics literature on state capacity, and it contributes to research on the politics of pandemics.

Importance: Many physicians believe that most medical malpractice claims are random events. This study assessed the association of prior paid claims, including a single prior claim, with future paid claims; whether public disclosure of prior paid claims affects future paid claims; and whether the association of prior and future paid claims decayed over time. Objective: To examine the association of 1 or more prior paid medical malpractice claims with future paid claims. Design, Setting, and Participants: This retrospective case-control study included all 881,876 licensed physicians in the US. All data analysis took place between July 2018 and January 2023. Exposure: Paid medical malpractice claims. Main Outcome and Measures: Association between a prior paid medical malpractice claim and likelihood of a paid claim in a future period, compared with simulated results expected if paid claims are random events. Using the same outcomes, we also assessed whether public disclosure of paid claims affects future paid claim rates. Results: Overall, 3.3% of the 841,961 physicians with 0 paid claims in the prior period had 1 or more claims in the future period vs 12.4% of the 34,512 physicians with 1 paid claim in the prior period; 22.4% of the 4,189 physicians with 2 paid claims in the prior period; and 37% of the 1,214 physicians with 3 paid claims in the prior period. The association between prior claims and future claims was similar for high-medical-malpractice-risk and lower-risk specialties. The predictive power of a prior paid claim for future claims declined gradually as the time since the prior claim increased, for prior or future periods up to 10 years. Public disclosure did not affect the association between prior and future paid claims. Conclusions and Relevance: In this study of paid medical malpractice claims for all US physicians, a single prior paid claim was associated with substantial, long-lived higher future claim risk, independent of whether a physician was practicing in a high- or low-risk specialty, or whether a state publicly disclosed paid claims. Timely, noncoercive intervention, including education, has the potential to reduce future claims.

In international politics, is environmental protection largely a "rich-country" priority? We perceive four reasons why, although individual exceptions are possible, the answer would be yes: as a country meets more of its basic economic needs, it can better take on environmental policy's long-term thinking, policy expenses, collective action problems, and quality-of-life issues. To cut through lip service paid by governments that are not serious about environmental protection, and the fact that the BASIC countries occupy a gray area between rich and poor, we employ computer-assisted textual analyses on all 3,774 paragraphs of statements made by national governments between 1995-2012 in the Committee on Trade and Environment within the World Trade Organization. Controlling for other factors, we find a general pattern of environmental discussions increasing as development level increases. This contributes substantively and methodologically to the literatures on the environment/development nexus and rising powers.

Big data has been fruitfully leveraged as a supplement for survey data - and sometimes as its replacement - and in the best of worlds, as a "force multiplier" to improve survey analytics and insight. We detail a use case, the big data classifier (BDC), as a replacement to the more traditional methods of targeting households in survey sampling for given specific household and personal attributes. Much like geographic targeting and the use of commercial vendor flags, we detail the ability of BDCs to predict the likelihood that any given household is, for example, one that contains a child or someone who is Hispanic. We specifically build 15 BDCs with the combined data from a large nationally representative probability-based panel and a range of big data from public and private sources, and then assess the effectiveness of these BDCs to successfully predict their range of predicted attributes across three large survey datasets. For each BDC and each data application, we compare the relative effectiveness of the BDCs against historical sample targeting techniques of geographic clustering and vendor flags. Overall, BDCs offer a modest improvement in their ability to target subpopulations. We find classes of predictions that are consistently more effective, and others where the BDCs are on par with vendor flagging, though always superior to geographic clustering. We present some of the relative strengths and weaknesses of BDCs as a new method to identify and subsequently sample low incidence and other populations.

How do procedural innovations, such as committee bypass, affect our roll-call-based measurements of individual member ideology - and therefore our measurements of polarization? Congressional polarization, measured using member ideal points derived from scaling roll call data, has been steadily increasing over the last half-century. However, changes in legislative procedure that affect the construction of the roll call record have been concurrent with this apparent increase in polarization. In this paper, we explore the effect of one unorthodox procedure - the use of committee bypass in the House - on the measurement of member ideology and chamber polarization. We utilize matching to generate balanced subsets containing similar bills that bypassed committee to reach the floor and bills that went through regular order. With these matched subsets, we estimate the effect of committee bypass on roll call votes and the resulting ideal points and polarization measures. We find that committee bypass has the effect of dampening, rather than exacerbating polarization.

IRT models are among the most commonly used latent trait models in all of political science, particularly in the estimation of ideal points of political actors in institutions. While widely used, IRT models are often misapplied, and a key element of their estimation, the item parameters, are almost always ignored and discarded. In this paper, we look into the application of IRT models to the estimation of judicial ideology scores by Martin and Quinn. Building off of a replication and extension of Martin and Quinn, we demonstrate that the often-ignored item parameters are, in fact, inconsistent with the assumptions of IRTs. Then, using a post-estimation fix that is designed to ameliorate the problem, we run the model again, generating new scores. We then compare our new ideal points to the existing ideal points and discuss the implications for both ideal point modeling generally and in judicial politics specifically. We conclude by replicating a prominent study in judicial politics that demonstrates how inconsistencies in the estimation of IRT models can be consequential and bring up concerns with the implications for what this could mean for the usefulness of scores estimated via IRT models.

We extend Studdert et al. We examine to what extent a physician who has past paid medical malpractice claims in a defined prior period is more likely to have additional paid claims in a defined future period, relative to a physician with no prior-period claims. Our simulation implements a null hypothesis that paid claims are random events, with arrival risk depending on state, but not on physician-specific factors such as technical ability, bedside manner, and communication skills. We show that even a single paid claim in the prior five years nearly quadruples the likelihood of a paid claim in the next five years, and dramatically increases the likelihood of 2+ future paid claims. More generally, the number of prior paid claims strongly predicts both the likelihood of having future paid claims and the expected number of future claims. By comparing actual to simulated probabilities, we can predict the likelihood that having a given number of paid claims is attributable to chance. We find that even for physicians in high-risk specialties in high-risk states, bad luck is highly unlikely to explain three or more claims in 5 years. Hospitals and state medical boards can use our approach to identify physicians that are likely to benefit from graduated interventions aimed at reducing future claims and patient harm.

Skirting the lines between academic, promotional and advocacy organizations, think tanks spend an inordinate amount of time and money attempting to influence policy debates, all the while being legally barred from lobbying. Think tanks, unlike interest groups, do not bring with them electoral constituencies to advocate on behalf of, so the ways in which they persuade legislators to adopt their opinions cannot simply be electoral in nature. Using a dataset of think tank citations from congressional floor speeches and committee testimony records, I compare the influence of think tanks based on a new measure of their ideologies and, in doing so, show that think tanks engage in strategic ideological positioning to maximize their influence. An additional hypothesis examined is the relationship between think tank members' previous work experiences in government with the organizations' overall prominence. By treating think tanks as strategic actors in legislative politics, I show that think tanks' ideological positioning affects directly how members of Congress engage with them, both by citing them in floor speeches and in calling them to testify, with more extreme think tanks being cited more frequently in floor speeches and more moderate think tanks called more often to testify.

Party leaders face a significant trade-off financing races when the party is out of power: while they care about gaining control of the House, they do not know how willing a potential representative will be to work with and for the party once elected. Leadership political action committee (LPAC) contributions are a major mechanism of leadership control over the financing of congressional campaigns, with the hope of influencing the future behavior of candidates. We study differences between contributions of the LPACs for leaders of both parties conditional on majority status. We find that both majority and minority party leaders prioritize winning elections and ideological homogeneity in their donations, but that these trends are largely contingent on overall electoral conditions. In their contributions, majority party leaders pay more attention to ideological cohesion than minority party leaders, while minority party leaders are more interested in gaining seats in the House than majority party leaders.

Evaluates direct LLM-based estimates of population quantities against Survey of Consumer Finances benchmarks, emphasizing substitution risk and the need for target-specific validation..

Develops and validates a survey-specific detector for AI-like open-ended responses as one signal in layered data-quality review, not as a standalone fraud determination..

Center for the Study of the Administrative State Working Paper series, George Mason Law.

Prepared for presentation at the 2020 Annual Meeting of the American Political Science Association.

This position paper argues that evaluating bias in AI agents remains a social-science measurement problem. It distinguishes dimensions of bias, separates observed bias from its causes, and highlights delegation as a source of bias in agentic systems.

This conference paper examines how machine learning and disambiguated author identifiers can improve record linkage in large-scale funding program evaluation. The goal is to reduce linkage error in publication and administrative data so that evaluation pipelines can recover research outputs and downstream impact more accurately.

An essay on James Q. Wilson's intellectual legacy and his place in political science.

A review essay on Timothy Groseclose's Left Turn and debates over media bias.