Sunday, Aug 3: 2:00 PM - 3:50 PM
4005
Contributed Papers
Music City Center
Room: CC-205C
Main Sponsor
International Chinese Statistical Association
Presentations
While strategic incentive policies are essential in personalized services, variations in user responses and revenue potentials create significant challenges in identifying optimal incentive policies. Existing literature typically falls short in fully identification of all user types and often assumes a uniform conversion revenue, leading to inefficient incentive allocations. In this study, we propose a minimax policy learning approach within a counterfactual principal strata framework. A value function, accommodating varying rewards across six potentially non-identifiable principal strata, is designed to minimize the worst-case value loss relative to three alternative policies: never-treat, always-treat, and oracle. To learn an optimal policy, we introduce three estimators: Principal Outcome Regression (P-OR), Principal Inverse Propensity Scoring (P-IPS), and Principal Doubly Robust (P-DR), providing theoretical guarantees for their unbiasedness, robustness, and regret upper bound. Extensive numerical experiments validate the effectiveness and superiority of our proposed approach.
Keywords
Policy learning
Minimax
Counterfactual principal strata
Partial identification
Causal inference
Co-Author(s)
Yue Liu, School of Statistics, Renmin University of China
Hao Mei, School of Statistics, Renmin University of China
First Author
Chenyang Li, School of Statistics, Renmin University of China
Presenting Author
Hao Mei, School of Statistics, Renmin University of China
This research explores the application of Functional Data Analysis (FDA) to uncover time-varying patterns in complex dynamic systems. FDA treats data as smooth, continuous functions rather than discrete observations, enabling the analysis of temporal dependencies and underlying dynamic structures. Building on this framework, we apply Bayesian Principal Component Analysis (BPCA) and Bayesian Factor Analysis (BFA) to enhance the study of time-dependent data. BPCA is extended to functional data to address challenges in dimensionality reduction for high-dimensional and noisy datasets, providing robust uncertainty quantification and improved interpretability of temporal principal components. Complementing this, BFA is applied at ten distinct time points, using smoothing splines to visualize dynamic factor loadings and scores, capturing the evolution of latent structures over time. Together, these Bayesian approaches advance our understanding of temporal correlations in dynamic systems, offering a flexible and detailed methodology for applications such as gene regulatory networks and other evolving processes.
Keywords
Functional Data Analysis (FDA)
Bayesian Principal Component Analysis (BPCA)
Bayesian Factor Analysis (BFA)
Temporal Dependencies
Dynamic Systems
Smoothing Splines
First Author
Boshi Zhao, Northern Illinois University
Presenting Author
Boshi Zhao, Northern Illinois University
Achieving covariate balance in randomized experiments enhances the precision of treatment effect estimation. However, existing methods often require heuristic adjustments based on domain knowledge and are primarily developed for binary treatments. This paper presents Gaussianized Design Optimization, a novel framework for optimally balancing covariates in experimental design. The core idea is to Gaussianize the treatment assignments: we model treatments as transformations of random variables drawn from a multivariate Gaussian distribution, converting the design problem into a nonlinear continuous optimization over Gaussian covariance matrices. Compared to existing methods, our approach offers significant flexibility in optimizing covariate balance across a diverse range of designs and covariate types. Adapting the Burer-Monteiro approach for solving semidefinite programs, we introduce first-order local algorithms for optimizing covariate balance, improving upon several widely used designs. Furthermore, we develop inferential procedures for constructing design-based confidence intervals under Gaussianization and extend the framework to accommodate continuous treatments. Simulations demonstrate the effectiveness of Gaussianization in multiple practical scenarios.
Keywords
Optimal Experimental Design
Covariate Balance
Continuous Treatments
Mehler's Formula
Traditional two-arm randomized trial designs have played a pivotal role in
establishing the efficacy of medical interventions. However, their efficiency is
often compromised when confronted with multiple experimental treatments
or limited resources. In response to these challenges, the multi-arm multi-stage
designs have emerged, enabling the simultaneous evaluation of multiple
treatments within a single trial. In such an approach, if an arm meets
efficacy success criteria at an interim stage, the whole trial stops and the arm
is selected for further study. However when multiple treatment arms are
active, stopping the trial at the moment one arm achieves success diminishes
the probability of selecting the best arm. To address this issue, we have
developed a group sequential multi-arm multi-stage survival trial design
with an arm-specific stopping rule. The proposed method controls the
familywise type I error in a strong sense and selects the best promising
treatment arm with a high probability.
Keywords
Multi-arm multi-stage
group sequential design
log-rank test
sequential conditional probability ratio test
time-to-event
Co-Author(s)
Yimei Li
Liang Zhu, St. Jude Children's Research Hospital
Tushar Patni, St Jude
First Author
Jianrong Wu, University of New Mexico, Dept. of Internal Medicine
Presenting Author
Jianrong Wu, University of New Mexico, Dept. of Internal Medicine
In meta-analysis, unlike model-based methods, such as fixed- or random-effect models, the p-value combining methods are distribution-free and robust. How to appropriately and powerfully combine p-values obtained from various sources remains an important but challenging topic in statistical inference. For cases where all or the majority of the individual alternative hypotheses have the same but unknown direction, concordant tests based on one-sided p-values can substantially improve the detecting power. However, there exist no uniformly most powerful tests; therefore, how to choose a robust and powerful test to combine one-sided p-values for a given data set is desirable. In this paper, we propose and study a class of gamma distribution-based concordant tests. Those concordant tests are optimal under specific conditions. An asymptotically optimal concordant test is also studied. The excellent performances of the proposed tests were demonstrated through numeric simulation study and real data example.
Keywords
chi-square test
constrained likelihood ratio test
gamma distribution
meta-analysis
uniformly most powerful test
Sensitivity analysis is essential for evaluating the robustness of causal conclusions in observational studies, particularly when key assumptions such as no unmeasured confounding and overlap may be violated. In this paper, we propose a robust marginal sensitivity model that accounts for the potential heterogeneity in unmeasured confounding strength and allows for non-overlap between treated and control units. We focus particularly on overlap-weighted average treatment effect that can better accommodate extreme treatment probabilities, and construct confidence intervals for the average treatment effect under given constraints on the violation of unconfoundedness and overlap. Our analysis utilizes constraints from the covariate balancing property of the true propensity score, and the proposed inference is applicable to general Z-estimation with overidentified equations and unmeasured variables.
Keywords
Causal inference
Sensitivity analysis
Unmeasured confounding
Non-Overlap
Overlap-weighted average treatment effect
Co-Author
Xinran Li
First Author
Han Cui, University of Illinois Urbana-Champaign
Presenting Author
Han Cui, University of Illinois Urbana-Champaign
Life history data describes a process that progresses through various stages before reaching a terminal event, providing detailed insights into the entire disease trajectory. It is increasingly used in health research to evaluate treatment effects, risk factors, and policy implementations. However, handling loss to follow-up (LTF) remains a major challenge since conventional multistate models often assume that LTF-induced censoring is independent of the life history process, an assumption that may not hold in practice. This paper addresses this issue through sensitivity analysis, which characterizes deviations from independent censoring and evaluates the impact on treatment effect estimation. We extend the classical multistate model to include separate pre- and post-LTF transition intensities. Using trace data informed sensitivity assumption, we apply a model-based multiple imputation (MI) to generate the entire latent transitions before and after LTF. We assess the performance of our method through simulation and apply the method using real-world data to assess the impact of the World Health Organization's Treat All policy on HIV disease progression.
Keywords
life history data
multistate model
lost to follow-up
multiple imputation.
sensitivity analysis