YSYoumi Suk

Youmi Suk

*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


Youmi Suk has a total of 73 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Fair and Robust Estimation of Heterogeneous Treatment Effects for Optimal Policies in Multilevel Studies

    by Youmi Suk, Chan Park, Chenguang Pan, Kwangho Kim

    In recent decades, shifting from the average treatment effect (ATE), researchers have turned to heterogeneous treatment effects or conditional average treatment effects (CATEs), especially using ma...

  • Rethinking Item Fairness Using Single World Intervention Graphs

    by Youmi Suk, Weicong Lyu

    Since the 1960s, the testing community has striven to ensure fairness in tests and test items. Differential item functioning (DIF) is a widely used statistical notion for items that may unfairly disadvantage specific subgroups of test-takers. However, traditional DIF analyses focus only on statistical relationships in observed data and cannot explain why such unfairness occurs. To fill this gap, we introduce a novel causal framework for defining and detecting unfair items using single-world intervention graphs (SWIGs). By leveraging SWIGs and potential outcomes, we define causal DIF (CDIF) as the difference in item functioning between two hypothetical worlds: one where individuals were assigned to one group and another where they were assigned to a different group. We also connect CDIF to related fairness concepts, including group versus individual fairness and item impact. In particular, we use SWIGs to graphically distinguish between item fairness at the individual level and the population level. Additionally, we discuss causal identification strategies using SWIGs and demonstrate how our approach differs from traditional DIF methods through a simulation study. We further illustrate its application to a controversial item from New York’s Regents math exam and a real dataset from the Trends in International Mathematics and Science Study, concluding with broader implications of promoting causal fairness in testing practices.


  • Open Access

    Causal Mediation and Functional Outcome Analysis with Process Data

    by Youmi Suk, Chan Park

    Over the past two decades, there has been growing interest in analyzing the effects of educational programs on outcomes using process data from computer-based testing and learning environments. However, most analyses focus on final outcomes at the end of a test or session, overlooking their functional nature over time and neglecting causal mechanisms. To address this gap, this article proposes a novel causal mediation framework for identifying and estimating functional natural direct effects, functional natural indirect effects, and functional total effects, along with their subgroup effects. We define these effects using potential outcomes and provide nonparametric identification strategies depending on whether post-treatment covariates are present or not. We then develop estimation methods using generalized additive models, a flexible and robust tool for analyzing functional data. Through a simulation study, we assess the finite-sample performance of the proposed approach by comparing it to parametric regression methods. We also demonstrate our approach by examining the effects of extended time accommodations on two functional outcomes using process data from the National Assessment of Educational Progress. Our mediation approach with functional outcomes effectively captures dynamic causal mechanisms underlying the program’s effects and pinpoints when and for whom each effect manifests throughout the testing period.


  • Open Access

    Identifying Causes of Test Unfairness: Manipulability and Separability

    by Youmi Suk, Weicong Lyu

    Differential item functioning (DIF) is a widely used statistical notion for identifying items that may disadvantage specific groups of test-takers. These groups are often defined by non-manipulable characteristics, e.g., gender, race/ethnicity, or English-language learner (ELL) status. While DIF can be framed as a causal fairness problem by treating group membership as the treatment variable, this invokes the long-standing controversy over the interpretation of causal effects for non-manipulable treatments. To better identify and interpret causal sources of DIF, this study leverages an interventionist approach using treatment decomposition proposed by Robins and Richardson (2010). Under this framework, we can decompose a non-manipulable treatment into intervening variables. For example, ELL status can be decomposed into English vocabulary unfamiliarity and classroom learning barriers, each of which influences the outcome through different causal pathways. We formally define separable DIF effects associated with these decomposed components, depending on the absence or presence of item impact, and provide causal identification strategies for each effect. We then apply the framework to biased test items in the SAT and Regents exams. We also provide formal detection methods using causal machine learning methods, namely causal forests and Bayesian additive regression trees, and demonstrate their performance through a simulation study. Finally, we discuss the implications of adopting interventionist approaches in educational testing practices.

  • Open Access

    Fair and Robust Estimation of Heterogeneous Treatment Effects for Optimal Policies in Multilevel Studies

    by Youmi Suk, Chan Park, Chenguang Pan, Kwangho Kim

    Recently, there have been growing efforts in developing fair algorithms for treatment effect estimation and optimal treatment recommendations to mitigate discriminatory biases against disadvantaged groups. While most of this work has focused on addressing discrimination due to individual-level sensitive variables (e.g., race/ethnicity), it overlooks the broader impact of societal structures and cultural norms (e.g., structural racism) beyond the individual level. In this paper, we formalize the concept of multilevel fairness for estimating heterogeneous treatment effects to improve fairness in optimal policies. Specifically, we propose a general framework for the estimation of conditional average treatment effects under multilevel fairness constraints that incorporate sensitive variables from multiple structural levels. Using this framework, we analyze the trade-off between fairness and the maximum achievable utility by the optimal policy. We evaluate the effectiveness of our framework through a simulation study and a real data study on advanced math courses using data from the High School Longitudinal Study of 2009.

  • Open Access

    Learning Feasible Optimal Treatment Regimes for Personalized Decision-Making

    by Chenguang Pan, Yuxuan Li, Youmi Suk

    Personalized decision-making is essential in psychology and education for addressing the specific needs, interests, and abilities of individuals. A popular, data-driven approach for personalized recommendations is based on optimal treatment regimes (OTRs), which finds a decision rule that maximizes a desirable mean outcome by leveraging treatment effect heterogeneity. However, standard OTRs may yield infeasible recommendations when institutional, developmental, or contextual factors restrict available options. This study introduces a general framework for feasible OTRs that explicitly integrates feasibility constraints into individualized decision rules in two ways: (a) a priori rules informed by domain knowledge (e.g., prerequisites) and (b) propensity-score-based trimming that excludes rarely observed treatment options. We conduct a series of simulation studies to systematically evaluate the performance of feasible OTR methods and compare Targeted Maximum Likelihood Estimation (TMLE) with Q-learning. Our first simulation study shows that the TMLE method performs comparably to an oracle Q-learning model when feasibility constraints are applied. The second simulation study identifies consistent ranges of optimal thresholds (0.01–0.08) across scenarios with varying numbers of treatment levels. We also develop a semi-synthetic calibration procedure to determine dataset-specific thresholds. Using data from the High School Longitudinal Study of 2009 (HSLS:09), we demonstrate the proposed framework by developing a personalized math course-taking recommendation model. The resulting feasible OTRs recommend personalized yet realistic course placements and increase the expected ninth-grade GPA relative to observed assignments. These results provide systematic evidence on incorporating feasibility into OTRs and offer practical guidance for designing realistic OTRs in personalized decision-making.

  • Using Generative AI for Sequential Data Generation in Monte Carlo Simulation Studies

    by Youmi Suk, Chenguang Pan, Ke Yang

    Monte Carlo simulation studies are the de facto standard for assessing both classical and modern statistical methods. In traditional simulation studies, researchers have full discretion over data generation and often create simulated data with a high degree of smoothness and limited dependencies among variables. Under such conditions, the performance of statistical methods may not indicate their real-world performance. To address this issue, we propose a novel AI-based simulation framework that leverages generative AI (GenAI) to create realistic synthetic data and incorporate them into Monte Carlo simulation studies. In particular, we focus on simulation studies with sequential data (e.g., action sequences from process data) and demonstrate our approach by evaluating the predictive performance of an example statistical method. Our framework consists of five key steps: (a) pre-processing input data, (b) training GenAI models on input data, (c) assessing synthetic data quality, (d) conducting AI-based simulations, and (e) evaluating simulation results. We also perform robustness checks on both synthetic data quality and simulation outcomes by modifying specific steps of our approach. Overall, the proposed AI-based simulations outperform traditional simulations in generating realistic synthetic data and providing a more accurate evaluation of the real-world performance of statistical methods.

  • Open Access

    Rethinking Item Fairness Using Single World Intervention Graphs

    by Youmi Suk, Weicong Lyu

    Since the 1960s, the testing community has striven to ensure fairness in tests and test items. Differential item functioning (DIF) is a widely used statistical notion for items that may unfairly disadvantage specific subgroups of test-takers. However, traditional DIF analyses focus only on statistical relationships in observed data and cannot explain why such unfairness occurs. To fill this gap, we introduce a novel causal framework for defining and detecting unfair items using single world intervention graphs (SWIGs). By leveraging SWIGs and potential outcomes, we define causal DIF (CDIF) as the difference in item functioning between two hypothetical worlds: one where individuals were assigned to one group and another where they were assigned to a different group. We also connect CDIF to related fairness concepts, including group versus individual fairness and item impact. In particular, we use SWIGs to graphically distinguish between item fairness at the individual level and the population level. Additionally, we discuss causal identification strategies using SWIGs and demonstrate how our approach differs from traditional DIF methods through a simulation study. We further illustrate its application to a controversial item from New York's Regents math exam and a real dataset from the Trends in International Mathematics and Science Study, concluding with broader implications of promoting causal fairness in testing practices.


  • Open Access

    Supplementary Figure from Relations of Current and Past Cancer with Severe Outcomes among 104,590 Hospitalized COVID-19 Patients: The COVID EHR Cohort at the University of Wisconsin

    by Margaret B. Nolan, Thomas M. Piasecki, Stevens S. Smith, Timothy B. Baker, Michael C. Fiore, Robert T. Adsit, Daniel M. Bolt, Karen L. Conner, Steven L. Bernstein, Oliver D. Eng, et al.

    Supplementary Figure from Relations of Current and Past Cancer with Severe Outcomes among 104,590 Hospitalized COVID-19 Patients: The COVID EHR Cohort at the University of Wisconsin