49755 research outputs found
Sort by
Building a Developmental Science of Redemption
Full text linkStories about redemption are ubiquitous; people emphasize moral improvement when describing their own lives and, often, others' lives as well. However, psychology does not yet have a well-developed literature concerning redemption, and developmental science has not addressed questions regarding how perceptions of redemption might emerge or change between childhood and adulthood. To the extent that past research has spoken to this issue, it has pointed in contradictory directions. Two different theories—focusing on essentialism and on optimism—make two different developmental predictions about how and why judgments of redemption might change with age. Integrating these perspectives, we propose a novel theory of redemption that puts work on essentialism and optimism in conversation with each other. The theory of redemption further highlights the role of social inputs (e.g., experiences with their own and others' moral change) as mechanisms that can lead children to hold more redemptive views than do adults. The theory of redemption accounts for previous findings in developmental science and makes novel predictions regarding the social inputs and consequences of redemptive views
Seasonal productivity of the equatorial Atlantic shaped by distinct wind-driven processes
No full textThe eastern equatorial Atlantic hosts a productive marine ecosystem that depends on upward supply of nitrate, the primary limiting nutrient in this region. The annual productivity peak, indicated by elevated surface chlorophyll levels, occurs in the Northern Hemisphere summer, roughly coinciding with strengthened easterly winds. For enhanced productivity in the equatorial Atlantic, nitrate-rich water must rise into the turbulent layer above the Equatorial Undercurrent. Using data from two trans-Atlantic equatorial surveys, along with extended time series from equatorial moorings, we demonstrate how three independent wind-driven processes shape the seasonality of equatorial Atlantic productivity: (1) the nitracline shoals in response to intensifying easterly winds; (2) the depth of the Equatorial Undercurrent core, defined by maximum eastward velocity, is controlled by an annual oscillation of basin-scale standing equatorial waves; and (3) mixing intensity in the shear zone above the Equatorial Undercurrent core is governed by local and instantaneous winds. The interplay of these three mechanisms shapes a unique seasonal cycle of nutrient supply and productivity in the equatorial Atlantic, with a productivity minimum in April
due to a shallow Equatorial Undercurrent and a productivity maximum in July resulting from a shallow nitracline coupled with enhanced mixing
Probabilistic machine learning for predictions and causal discovery in health informatics.
No full textThis thesis is about probabilistic machine learning methods for predictive modeling, causal discovery, and applications in health informatics. Probabilistic models have achieved remarkable performance in text and image modeling, often predicting answers to textual or visual questions better than human experts. But can these methods succeed as well on health data questions? Health data are not limited to text or images -- they include tabular datasets from electronic health records, high-dimensional genomic measurements from single-cell sequencing, or time series from wearable devices. Health questions are not only about predicting an answer alone -- they involve explaining those predictions and discovering unknown relations between variables. This thesis addresses these challenges. It introduces methods that improve probabilistic prediction on common health data types and scale causal discovery to the thousands of variables encountered in genomic datasets. It then develops interpretable generative models for concrete health applications like cancer research with single-cell gene expression and heart health research with wearable time series. These methods are designed for non-experts in machine learning, with software implementations that require little to no tuning.
The thesis is organized into four parts, with the first two focusing on broadly applicable methodologies and the last two on concrete health applications of the previous methods. The first part focuses on probabilistic predictions, the task of estimating the full distribution of a target variable given other variables. This task is critical in health applications to quantify uncertainty, compute risk, or detect anomalies. We introduce two methods: treeffuser and unbounded depth neural networks (UDNs).Treeffuser provides probabilistic predictions for tabular data using a diffusion model parameterized by gradient-boosted trees.
In contrast, UDNs are deep neural networks. They provide probabilistic predictions as a mixture of outputs from all of their hidden layers. Importantly, UDNs automatically adapt their depth to the complexity of the data during training.
Both methods require minimal tuning and improve on existing methods.
The second part of this thesis focuses on causal discovery, the task of inferring causal relationships between variables. It is a fundamental task in health science, but existing methods hardly scale to the hundreds of variables of modern datasets. We develop two scalable methods: extreme greedy equivalence search (XGES) and stable differentiable causal discovery (SDCD). XGES is designed for linear models and has provable guarantees, whereas SDCD is designed for neural networks. Both methods improve convergence speed and accuracy, enabling causal discovery to scale to thousands of variables.
The third part designs probabilistic models in single-cell genomics. Single-cell RNA sequencing (scRNA-seq) measures gene expression across thousands of cells. But scRNA-seq data is challenging to analyze and usually requires multiple steps that can fail: batch correction, dimensionality reduction, data visualization, trajectory analysis, and gene pattern analysis. We propose Decipher, a tool for analyzing single-cell data that unifies all those steps and addresses their limitations. Decipher is a deep generative model. It learns a low-dimensional representation of each cell's state along with a two-dimensional visualization. Incorporating the visualization within Decipher's model enables new types of trajectory and gene pattern analyses. Applied to acute myeloid leukemia data, Decipher successfully maps the divergence from normal hematopoiesis and identifies transcriptional programs associated with NPM1 mutations.
The fourth and last part focuses on physiological time-series data recorded by the Apple Watch. We demonstrate through two studies how probabilistic models can uncover insights into fitness, heart rate regulation, and changes in human behavior. The first study models the subjects' heart rate given their activity intensity measured via GPS speed and step count. The study builds on existing physiological heart models based on differential equations and augments them with probabilistic machine-learning components. The resulting model forecasts heart rate responses better than standard deep learning models, learns personalized fitness indicators, and reveals how much environmental factors impact heart rate. The second study estimates the causal effect of the Apple Watch ``time to stand'' reminder using a regression discontinuity design specially adapted for time series. Using billions of minutes of standing data, it discovers that the nudge increases standing rates by up to 43.9\% and that it remains effective over time.
With this thesis, health researchers gain tools to uncover deeper insights into the human body, and machine learning practitioners gain methodologies for developing such tools on complex health data
Japanese Colonial Architecture in South Korea: Changes in Perception from 1945-2025
No full textThis thesis explores the changing perceptions surrounding the preservation of Japanese colonial architecture in South Korea. By examining key developments in the country’s preservation field over the past century, the study examines how attitudes toward Japanese colonial architecture have changed since 1945, framing this evolution within the broader development of preservation practices in South Korea. The thesis finds that the reasons for preserving these buildings are multifaceted, as demonstrated through the analysis of three case study sites. However, a consistent theme across all cases is the role of Korean national identity and historical significance in shaping the rationale for preserving colonial-era architecture
Problems and Prospects of Democratic Design with Roundtable Discussion
No full textThis project investigates the relationship between participation patterns—namely, ones associated with equality and variety—in roundtable discussions and teachers’ perceptions of their democratic quality, and it does so through an integrative methodological approach. While discussion-based learning tends to promote democratic pedagogical practice, a limited amount of empirical research examines how the structural and content features of discussions, such as turn-taking patterns, influence teachers’ evaluations of their deliberative and equitable nature.
The empirical chapters analyze data from school observations collected via Dialogic, a web-based program that tracks discussion dynamics, alongside on-site observations and semi-structured interviews with seven teachers. Phase I employs regression analyses and multilevel modeling to explore how proxy variables for variety and equality, as well as controls, predict teachers’ democracy scores. Phase II uses inductive thematic analysis and descriptive quantification to examine how teachers perceive and facilitate discussions.
Results suggest that increased variety and equality in turn-taking positively correlate with higher democracy scores, though these factors alone explain limited variance. Furthermore, smaller class sizes brought with them lower perceived democratic quality. Phase II reveals that teachers often act as counter-majoritarian facilitators, suggesting that optimal discussions blend democratic participation with strategic teacher intervention. The study concludes that roundtable discussions function best as a “mixed-rule” format, balancing student-driven dialogue with teacher guidance to foster equitable and deliberative learning spaces. These findings contribute to understanding how structural and facilitative elements shape democratic educational practices
Essays in Political Economy
No full textThis dissertation is a collection of five essays that examine the forces shaping contemporary U.S. political competition, with a focus on voter realignment, party strategy, policy diffusion, and democratic attitudes. It traces how shifts in voter preferences, party positioning, and institutional dynamics have reconfigured electoral coalitions, shaped legislative developments, and influenced political responses to major shocks, such as climate change and the Covid-19 crisis. The analysis draws on theory, newly assembled historical and contemporary datasets, survey experiments, and both reduced-form and structural estimation to uncover key mechanisms in political economy.
The first chapter estimates a political equilibrium model to disentangle demand factors (voters) from supply factors (politicians) in shaping political outcomes, focusing on the recent realignment of blue-collar voters away from left-wing parties. I jointly evaluate the impact of changes in voter preferences and voter demographics (demand side) and party positions and party discipline (supply side) on voters’ partisan realignment in U.S. House elections between 2000 and 2020. To measure candidate ideological positioning, I estimate a multimodal text-and-survey model from campaign websites. To estimate voter preferences, I build a new panel of precinct-level election results (N=1.3 million), which allows me to exploit congressional districts' border discontinuities for identification. The paper ultimately identifies parties’ stronger polarization on cultural issues compared to economic issues as the main driver of voters’ partisan realignment. In contrast, shifts in voter preferences—particularly the increasing preferences of blue-collar voters for progressive economic policies—have mitigated their defection from the Democratic Party. Absent these demand-side changes, voters’ partisan realignment would have been even more pronounced. Within specific policy domains, the environment emerges as the topic where parties diverge most in economic versus cultural emphasis: Democrats frame it culturally, while Republicans focus on economic aspects. Simulations reveal that a progressive, economically focused environmental policy would gain greater blue-collar voter support than a culturally focused one.
The second chapter examines the same question from a historical perspective, tracing partisan realignment by education back to World War II. It argues that the Democratic Party’s evolution on economic policy helps explain partisan realignment by education. We show that less-educated Americans differentially demand ``predistribution’’ policies (e.g., a federal jobs guarantee, higher minimum wages, protectionism, and stronger unions), while more-educated Americans differentially favor redistribution (taxes and transfers). This educational gradient in policy preferences has been largely unchanged since the 1940s. We then show the Democrats’ supply of predistribution has declined since the 1970s. We tie this decline to the rise of a self-described ``New Democrat’' party faction who court more educated voters and are explicitly skeptical of predistribution. Consistent with this faction’s growing influence, we document the significant growth of donations from highly educated donors, especially from out-of-district, who play an increasingly important role in Democratic (especially ``New Democrat'') primary campaigns relative to Republican primaries. In response to these within-party changes in power, less-educated Americans began to leave the Democratic Party in the 1970s, after decades of serving as the party’s base. Roughly half of the total shift can be explained by their changing views of the parties’ economic policies. We also show that in the crucial transition period of the 1970s and 1980s, New Democrat-aligned candidates draw disproportionately from more-educated voters in both survey questions and actual Congressional elections.
The third chapter estimates an empirical model of adoption and diffusion of policies across U.S. states since 1787. We use a large language model to extract and structure the policy content of the universe of statutes enacted by state legislatures (N=2.5 million). We then construct sets of related policies across states using an unsupervised clustering algorithm applied to vector representations of the statutes' policy contents. We compute measures of similarity, diffusion, and innovation across states with legislative activity on these policy clusters. After validating this measurement procedure, we analyze the determinants of diffusion. We find that states with greater geographic, economic, and political similarity implement more similar policies. While polarization has risen significantly since the 1990s, consistent with prior literature, we show that current levels of polarization are comparable to those observed in the pre-WWII period, following a U-shaped pattern over time. Polarization primarily reflects differences in policy choices within topics rather than in the topics on which states choose to legislate. Finally, we document an increasing nationalization of policy, with federal legislative texts exerting growing influence on state legislation since the post-war period.
The fourth chapter studies the effects of climate change and mitigation on U.S. politics. We combine 2000-2020 precinct-level voting information and congressional candidate positions on environmental policy with high-resolution temperature and precipitation data and census block-group level measures of ``green" and ``brown" employment shares. Holding politician positions fixed within a district, we find that Democratic vote share increases with exogenous changes in local climate and green transition employment. We embed these estimates into a structural model of political competition, including both direct and demand-driven effects of shocks on candidate policy platform supply. Incorporating our model estimates into 2025-2050 projections of climate change and green employment transition, we find that voting for the Democrats increases, while both parties move slightly to the right on climate policy. Under worst-case climate projections and current mitigation trajectories, the median 2050 Congressperson has roughly the same environmental ideology as the median 2010 Democrat---for instance supporting carbon pricing.
The fifth chapter studies how crisis of the magnitude of the Covid-19 pandemic may plausibly affect deep-seated attitudes of a large fraction of citizens. In particular, outcome-oriented theories imply that leaders' performance in response to such adverse events shapes people’s views about the government and about democracy. To assess these causal linkages empirically, we use a pre-registered survey experiment covering 12 countries and 22,500 respondents during the pandemic. Our design enables us to leverage exogenous variation in evaluations of policies and leaders with an instrumental variables strategy. We find that people use information on both health and economic performance when evaluating the government. In turn, dissatisfaction with the government decreases satisfaction with how democracy works, but it does not increase support for non-democratic alternatives. The results suggests that comparatively bad government performance mainly spurs internal critiques of democracy
Essays in Computational Law and Economics
No full textLegal institutions are surrounded by, constituted in and expressed through text. Recent advancements in natural language processing (NLP) techniques therefore hold great promise in expanding the scope of questions addressable by empirical law and economics research.
This dissertation demonstrates that promise by addressing three distinct questions in a manner that utilizes economic theory frameworks to answer questions that are grounded at some level in legal text. Specifically, this work explores: (1) how slavery affected technological innovation trajectories in the American South as revealed through patents, (2) how judicial experience and court structure impact securities litigation outcomes discernible in part through court filings and (3) how researchers can quantify and correct misclassification errors that arise when processing textual data such as legal opinions or merger-and-acquisition contracts.
This dissertation employs diverse NLP methodologies to extract meaningful information from legal documents across all three studies. The first paper applies Multinomial Inverse Regression (MNIR) and Large Language Models (LLMs) to analyze historical U.S. patent texts (1836-1877), combining these insights with a directed technical change model and difference-in-differences estimation.
The second paper uses a Random Forest Classification algorithm to process securities class action complaints and judicial opinions, developing a structural model of expert decision-making to evaluate judicial performance.
The third paper presents statistical frameworks for quantifying and correcting misclassification errors that occur when LLMs or humans classify features in legal texts, with validation through Monte Carlo simulations and empirical applications to legal data. The findings of these three studies reveal that institutional structures significantly shape economic outcomes across multiple domains.
The first study concludes that slavery directed Southern technological innovation away from capital-intensive production techniques, potentially explaining divergent industrial development between North and South. The second study demonstrates that judicial experience increases skill in identifying non-meritorious securities lawsuits while advanced age diminishes it, estimating that optimized term limits and specialized courts could have prevented over $14.5 billion in non-meritorious settlements since 1995. The third study provides practical methods for correcting both attenuation and directional bias from misclassification error in textual analysis.
Collectively, these studies illustrate how modern NLP tools can substantially expand the scope and precision of empirical law and economics research, unlocking new insights into the relationship between legal institutions and economic outcomes that were previously constrained by data limitations
Metal, Mylar, and Mirrors: On the Significance of the Interiors of Kevin Roche
No full textAre Late Modern and Postmodern commercial interiors worthy of preservation? How can a better understanding of these spaces help substantiate their significance as heritage? In light of increasing threats of renovation and demolition, this thesis explores these questions through the architectural interiors of Irish-American architect Kevin Roche, a prominent figure in Late Modern and Postmodern design whose interior work has recently faced preservation challenges. Contributing to ongoing discourse on the conservation of architecture from this era, the thesis focuses on Roche’s use of mirrored surfaces in the 1970s and 1980s, situating these design choices within the broader context of Postmodernism.
To closely examine Roche’s work, this study traces his design evolution and systematic architectural approach, drawing on published literature and an interview with one of his former principals. It also considers concurrent cultural and labor shifts in the United States that shaped the work of Roche and his contemporaries. These themes are explored through case studies of three key buildings, supported by primary source material from the Kevin Roche John Dinkeloo & Associates archive at Yale University.
Spurred by the recent demolition of one of Roche’s late-1980s buildings, this thesis argues that his later work is equally—if not more—significant than his earlier projects. His prolific use of mirrored finishes during this period reveals a thoughtful response to shifting material conditions, environmental concerns, and urban constraints, underscoring the lasting value of these interiors in the architectural heritage canon
Passive Cooling Revisited: Assessing the Decarbonization Potential of Window Awnings for Existing Buildings in New York City
No full textThis thesis investigates the overlooked potential of window awnings as a passive cooling strategy for historic and older buildings amid accelerating climate change and the rise in both global temperatures and air conditioning use. The research reevaluates awnings not as a decorative attachment for thermal comfort, but as an historically-grounded technology that holds a renewed relevance for operational carbon reductions today. With an interdisciplinary approach that integrates architectural history, building science, and three-dimensional computer energy modeling, a feasibility study evaluates the impact of a contemporary awning installation for a Harlem, New York apartment house constructed in 1901.
Energy modeling calculations demonstrate a potential savings of 98 metric tons of carbon dioxide equivalent over a 15 year period. At scale, estimated across even a modest share of New York’s historic building stock, these savings represent a significant operational carbon reduction opportunity. This thesis argues that historic preservationists are uniquely positioned to champion the strategic reintroduction of awnings through both awareness and policy reform, positioning them not as nostalgic embellishments, but as valuable tools for passive cooling, energy efficiency, and decarbonization of the existing built environment
Assessing the Impact of European Initiatives on Mitigating Severe Climate Effects in Egaleo: A Technological and Social Innovation Perspective
No full textThe Municipality of Egaleo is a densely inhabited city that is grappling with issues exacerbated by climate change. These challenges include an increase in vulnerable social groups, the presence of severe weather events, and a lack of preparedness for environmental disasters. This case study highlights the efforts and learning milestones of the municipality of Egaleo from implementing European-funded Sustainability Projects as solutions for tackling societal and environmental issues. It encapsulates the overcoming of various obstacles present in local and regional levels and how Egaleo is rapidly evolving as a European-level testbed for scalable, adaptive socio-environmental solutions.
Keywords: Climate change, risks, vulnerable groups, transformational adaptation, awarenes