SFSam Fuller

Sam Fuller

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Sam Fuller has a total of 18 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Beyond Common Spaces: Contrastive Scaling for Cross-Survey Structural Comparison

    by Tzu-Ping Liu, Sam Fuller

    This paper develops contrastive scaling as a framework for cross-survey structural comparisons. Whereas conventional bridging estimates a latent space shared across surveys, contrastive scaling identifies dimensions along which the organization of responses differs across them. We illustrate its utility using the UTokyo-Asahi Survey and a joint U.S.-U.K. dataset. The results show that, over time, Japanese voters have become increasingly divided over issues related to the status of the Jieitai (Japan Self-Defense Force), particularly in the structure of response distributions. In cross-national comparisons, U.S. and U.K. voters differ primarily along country-specific issue dimensions—such as attitudes toward Trump and Brexit—rather than along a common ideological axis. These findings demonstrate that contrastive scaling provides a complementary perspective on survey comparison by identifying dimensions of differential variation, offering a new lens for studying how public opinion is structured across contexts.

  • Open Access

    Unbundling Support for Political Violence

    by Sam Fuller, Jack T. Rametta, Alexa Federice

    Measures of support for political violence (SPV) vary drastically across what violence they describe, who performs it, against whom, and the stated justification. We argue that these measures are not interchangeable operationalizations of a single latent attitude but rather represent correlated, distinct subdimensions of SPV. Across three studies, we iteratively develop and validate concise batteries that 1) differentiate between actor-target combinations and reactive versus proactive violence and 2) vary across justification and the response's severity. We find that SPV is a multidimensional concept: levels of reactive support are remarkably higher than proactive; support for state-on-mass and mass-on-state violence are only weakly correlated, with Republicans supporting state-on-mass and Democrats supporting mass-on-state violence at significantly higher rates; common SPV measures correlate with proactive more than reactive; and we find mixed evidence that survey disengagement influences SPV. Our findings contextualize disparate and conflicting literatures, offering measurement strategies tailored to specific research questions.

  • Open Access

    Are Random Forests Still ‘Good Enough’? Tabular Prior-Data Fitted Networks for Predictive and Causal Tasks

    by Jack T. Rametta, Sam Fuller

    Are random forests, the workhorse of supervised machine learning methods in the social sciences, still “good enough” versus new methods that tout dramatic performance benefits? In this article we present a large, diverse Monte Carlo study and existing real-world data benchmarks to compare tree-based methods with TabPFN, a new pretrained transformer foundation model that performs in-context learning over millions of synthetic datasets designed for tabular data. We compare tree-based methods with TabPFN in the popular R-learner framework for conditional average treatment effect estimation. This allows us to assess both predictive model performance and resulting gains for downstream inference. First, our simulations suggest that TabPFN does outperform random forest, achieving near-oracle results for conditional effect estimation. TabPFN’s improvements manifest in the most difficult simulation setups, where the data generating process is complex and there are fewer observations. Second, in real-world data analyses TabPFN performs well, outperforming random forest in some cases, especially as sample dwindles and the number of predictive covariates increases. Our results suggest that tree-based methods are still well suited for social science data, but TabPFN specifically, and the prior data fitted network approach generally, is a strong competitor worthy of consideration.

  • Open Access

    Policy or Partisanship? How Polarization Biases Valence Evaluations in the U.S., U.K., and Taiwan

    by Sam Fuller, Tzu-Ping Liu

    How does affective and political polarization influence voters' evaluations of politicians? To answer this question, we use causal mediation analysis to examine three conjoint experiments fielded in the United States, the United Kingdom, and Taiwan that elicit respondents' evaluations of politicians' trustworthiness and governing competence. We find that political disagreement—differences in partisanship and/or policy preferences—influences evaluations, but the existence of direct effects and the size of total effects depends on whether partisanship serves as the dominant political social identity. In the U.S. and Taiwan, where partisanship serves as a clear social identity, co-partisanship has effects comparable to policy disagreement and exerts both direct and affect-mediated effects; whereas in the U.K., co-partisanship has weaker effects than policy disagreement, and only indirectly. These findings suggest that polarization influences valence evaluations generally, but operates either through partisanship or policy (dis)agreements depending upon a country's particular structure of political identities.

  • Open Access

    The Changing Landscape of Democratic (Dis)Satisfaction: Results from the American National Election Study 1996–2024

    by Sam Fuller, Neil Williams, Jack T. Rametta

    What predicts democratic satisfaction in the United States? Using the 1996–2024 time series as well as the 2000–02–04 and 2016–20–24 panels from the American National Election Study, this article identifies the consistent and time-specific predictors of democratic satisfaction. We analyze satisfaction using 1) random forests to identify important predictors; 2) causal forests to test established predictors; and 3) doubly-robust difference-in-differences to identify the causal effect of winning elections. We find that 1) predictors have shifted significantly over time—affect toward big business dominates after 2016—but election efficacy and government responsiveness are consistently predictive; 2) established predictors, except affective polarization and participation, are all significant; and 3) winning elections significantly increases satisfaction, and effects in 2004 are double the size of the midterm and general elections in 2002 and 2020, respectively. These findings contextualize the United States in the broader comparative literature, opening avenues for future research on democratic satisfaction.

  • Open Access

    Leaving Money on the Table: A Monte-Carlo Study Comparing Causal Forest and Standard Regression Models for Experiments

    by Sam Fuller, Jack T. Rametta

    Causal machine learning methods for experiments have proliferated, but adoption in political science is scarce. Are political scientists "leaving money on the table" by neglecting to apply these methods? To answer this question, we use over two million Monte-Carlo simulations to compare causal forest (CF) to standard regression techniques commonly used to estimate average and conditional average treatment effects. We find that CF outperforms standard methods in terms of statistical power, coverage, precision, bias, and the accurate detection and consequent estimation of heterogeneous treatment effects. However, CF underperforms in edge cases where measured covariates are entirely uninformative and sample sizes are very small. Moreover, we find that standard approaches have large false-positive rates for conditional treatment effects: 9–24 times larger than rates for CF. Our results suggest that researchers who choose to use CF stand to benefit greatly across statistical power, coverage, precision, bias, and heterogeneity detection.

  • Open Access

    What Predicts Support for Political Violence? Results from a Machine Learning Meta-Reanalysis

    by Sam Fuller, Jack T. Rametta, Alexa Federice

    There is significant research on support for political violence (SPV), its correlates, and interventions to reduce it. However, there is no framework for understanding who supports political violence. This article leverages the exploratory power of (causal) machine learning to conduct a registered meta-reanalysis of both the observational and experimental literature on SPV from 1995-2025 to synthesize and summarize previous research. First, we identify three categories of individual-level predictors: politically salient identities, psychological characteristics, and attitudes toward one's political system. Second, we find that young adults are the most likely to support political violence across many recent surveys, particularly in the U.S. Additionally, in two experiments in the U.S. and India we find that young people are by far the most affected by treatments aimed at both decreasing and increasing support, respectively. Third, and in contrast to research on the perpetration of violence, we find scant evidence of a relationship between gender identity and SPV.

  • Open Access

    Causal Forest and Doubly Robust Machine Learning for Political Science

    by Sam Fuller, Jack T. Rametta

    Machine learning (ML) methods in general and tree-based methods in particular are on the rise in the social sciences. While ML is widely valued for predictive accuracy, new causally-oriented machine learning methods actually leverage this power for statistical inference and treatment effect estimation. In this article, we synthesize this causal ML literature into a new paradigm we coin "Doubly Robust Machine Learning," or DRML. Our conceptual framework makes clear the benefits of these new methods, like causal forest, and explains how researchers can fruitfully employ these approaches to improve on the power, precision, and coverage of standard estimators. We demonstrate these benefits for average and conditional effect estimation through a massive array of over one million Monte Carlo simulations spanning a wide range of possible research settings. Taken together, our results provide the conceptual and empirical basis for the use of DRML in social science applications.

  • Open Access

    Attitudes Surrounding Fairness and Competition in Sports Predict Choices to Partisan Gerrymander

    by Hilary Jan Izatt, Sam Fuller

    Partisan polarization in the United States has intensified, fueling hostility toward partisan out-groups and eroding political and social trust. This divide has often been compared to the fervent loyalty of sports fans, where competition and “team spirit” dominate behaviors. Despite this comparison, research has not systematically explored how views on fairness and competitiveness in partisan competition predict support for anti-democratic policies. This paper addresses this gap by developing a novel survey battery that uses sports as a conceptual proxy to measure these attitudes. We test this survey battery, and further refine it, on two U.S. samples. Using dimensional analysis we recover two latent dimensions: fairness and competitiveness. Using a novel measure, a gerrymandering map-choice, we find that these dimensions are highly predictive of anti-democratic behavior. This study illustrates how individuals’ partisanship and underlying psychology lead to undemocratic outcomes in the context of partisan competition.


  • Open Access

    Attitudes Surrounding Fairness and Competition in Sports Predict Choices to Partisan Gerrymander

    by Hilary Izatt, Sam Fuller

    Partisan polarization in the United States has intensified, fueling hostility toward partisan out-groups and eroding political and social trust. This divide has often been compared to the fervent loyalty of sports fans, where competition and "team spirit" dominate behaviors. Despite this comparison, research has not systematically explored how views on fairness and competitiveness in partisan competition predict support for anti-democratic policies. This paper addresses this gap by developing a novel survey battery, grounded in social identity theory, that uses sports as a conceptual proxy to measure these attitudes. We test this survey battery, and further refine it, on two U.S. samples. Using dimensional analysis we recover two latent dimensions: fairness and the competitiveness. Using a novel measure—a gerrymandering map-choice—we find that these dimensions are highly predictive of anti-democratic behavior. This study illustrates how individuals' partisanship and underlying psychology lead to undemocratic outcomes in the context of partisan competition.