Data Fusion and Subgroup Analysis

Anamitra Chaudhuri Chair
The University of Texas at Austin
 
Monday, Aug 3: 10:30 AM - 12:20 PM
6262 
Contributed Papers 
Thomas M. Menino Convention & Exhibition Center 
Room: CC-106 

Main Sponsor

Section on Statistical Learning and Data Science

Presentations

Cross-fitting in causal machine learning with correlated units

In causal machine learning, the fitting and evaluation of nuisance models are typically performed on separate partitions, or folds, of the observed data. This technique, called cross-fitting, eliminates bias introduced by the use of black-box predictive algorithms. When study units may be correlated, such as in spatial, clustered, or time-series data, investigators often design bespoke forms of cross-fitting to minimize correlation between folds. We prove that, perhaps contrary to popular belief, this is typically unnecessary: performing cross-fitting as if study units were independent usually still eliminates key bias terms even when units may be correlated. In simulation experiments with various correlation structures, we show that causal machine learning estimators typically have the same or improved bias and precision under cross-fitting that ignores correlation compared to techniques striving to eliminate correlation between folds. 

Keywords

cross-fitting

causal machine learning

correlated data

spatial data

clustered data

semi-parametric estimation 

Speaker

Salvador Balkus, Harvard T.H. Chan School of Public Health

Co-Author(s)

Hasan Laith, Harvard College
Nima Hejazi, Harvard T.H. Chan School of Public Health

Divide-and-shrink: An efficient and heterogeneity-agnostic approach for transfer estimation

Knowledge transfer across data sources holds great promise for improving the estimation of target population parameters by leveraging the growing availability of data from different sources. However, the effectiveness of knowledge transfer is often challenged by the complex and pervasive heterogeneity between data sources and the lack of access to individual-level data. This paper proposes the divide-and-shrink (dShrink) method, a transfer estimation method that estimates target population parameters in a closed form using summary statistics from a target population and an external source population while accounting for population heterogeneity. dShrink is guaranteed to improve the estimator using the target population in the expected quadratic error under arbitrary population heterogeneity. Notably, it is model-free, requires no user-specified tuning parameters, robust to various types of heterogeneity between data sources, and applies to a broad range of parameter estimation problems. Simulations and real data analyses demonstrate the superior performance of the dShrink estimator and its potential as a robust tool for transfer estimation. 

Keywords

Combining information

Data integration

Efficiency

Population heterogeneity

Robustness

Shrinkage estimators 

Speaker

Ruoyu Wang, Harvard University

Co-Author

Xihong Lin, Harvard T.H. Chan School of Public Health

Efficient Subgroup Analysis via Optimal Trees with Global Parameter Fusion

Identifying and making statistical inferences on differential treatment effects-commonly known as subgroup analysis in clinical research-is central to precision health. Tree-based recursive partitioning methods are widely used for subgroup analysis due to their interpretability. Nevertheless, these approaches encounter significant limitations, including suboptimal partitions induced by greedy heuristics and overfitting from locally estimated splits. To address these limitations, we propose a fused optimal causal tree method that leverages mixed-integer optimization to facilitate precise subgroup identification. Our approach ensures globally optimal partitions and introduces a parameter-fusion constraint to facilitate information sharing across related subgroups. We provide theoretical guarantees by rigorously establishing out-of-sample risk bounds and comparing them with those of classical tree-based methods. Empirically, our method consistently outperforms popular baselines in simulations. We demonstrate its practical utility through a case study on the Health and Aging Brain Study–Health Disparities dataset, where our approach yields clinically meaningful insights. 

Keywords

Subgroup Analysis

Mixed integer optimization

Parameter Fusion

Optimal Tree 

Speaker

Zhongming Xie

Co-Author(s)

Josep Giorgio, University of California, Berkeley
Jingshen Wang, UC Berkeley

Estimation of causal dose-response functions under data fusion

Estimating the causal dose-response function is challenging, particularly when data from a single source are insufficient to estimate responses precisely across all exposure levels. To overcome this limitation, we propose a data fusion framework that leverages multiple data sources that are partially aligned with the target distribution. Specifically, we derive a Neyman-orthogonal loss function tailored for estimating the dose-response function within data fusion settings. To improve computational efficiency, we propose a stochastic approximation that retains orthogonality. We apply kernel ridge regression with this approximation, which provides closed-form estimators. Our theoretical analysis demonstrates that incorporating additional data sources yields tighter finite-sample regret bounds and improved worst-case performance, as confirmed via minimax lower bound comparison. Simulation studies validate the practical advantages of our approach, showing improved estimation accuracy when employing data fusion. This study highlights the potential of data fusion for estimating non-smooth parameters such as causal dose-response functions. 

Keywords

Data fusion

Orthogonal statistical learning

Causal dose-response function

Minimax

Reproducing kernel Hilbert space 

Speaker

Jaewon Lim

Co-Author

Alex Luedtke

Geometric Stochastic Interventions for Efficient Policy Learning

We introduce a geometric framework for constrained policy optimization using stochastic intervention analysis. Our goal is to estimate the causal effect of a policy that evolves smoothly from a behavioral baseline to an optimal target. We formulate the optimization problem as maximizing expected reward, subject to a Bregman divergence constraint and then show that the optimal solution lies on an e-geodesic, generated by a strictly convex non-affine generator G. For the expected reward along the geodesic path, an Efficient Influence function (EIF) is derived and an explicit decompositation of its variance is provided. We also develop a data-driven methodology to learn the generator G which minimizes the variance of EIF. This provides  a statistically efficient estimator, that extends beyond the limitations of exponentially family distributions. 

Keywords

Stochastic Interventions

Causal Inference

Constrained Policy Optimization

Efficient Influence Function 

Speaker

Debarghya Jana

Co-Author(s)

Kyle Schindl
Jae-Kwang Kim, Iowa State University

Robust Subgroup Analysis for Heterogeneous Censored Data

Subgroup analysis is important in practice because real-world data typically come from heterogeneous populations, where meaningful patterns can differ substantially across subpopulations. Correctly identifying these subgroups can improve prediction accuracy, prevent biased or misleading conclusions, and support more effective, targeted decision-making. While most existing subgroup analysis methods are developed for complete data, in this paper we propose a novel and robust approach for censored data under heterogeneous accelerated failure time (AFT) models. Specifically, we combine inverse probability weighting, M-estimation, and concave pairwise fusion penalization to simultaneously identify subgroups and estimate covariate effects for heterogeneous censored data, without requiring prior knowledge of individual subgroup memberships. We further develop an efficient RISA-ADMM algorithm to implement the method and establish its convergence. Furthermore, we derive the theoretical properties of the proposed estimators under mild regularity conditions. Extensive simulations and an application to the German credit dataset demonstrate the robustness and effectiveness of our approach. 

Keywords

Accelerated failure time model

Censored outcomes

Fusion penalization

Heterogeneity

Inverse probability weighting

Subgroup identification 

Speaker

Daoji Li, California State University, Fullerton

Co-Author(s)

Zhaohui Xu, University of Science and Technology of China
Zemin Zheng, University of Science and Technology of China

Statistical Benchmarking of Domain-Adapted Transformers for Financial Text Sentiment

Sentiment analysis in financial news presents statistical challenges due to high-dimensional lexical features and domain-specific polarity, where common terms deviate from general-usage sentiment. In this study, we present a comparative framework for three-class sentiment classification, benchmarking six model families: classical linear and non-linear learners (Logistic Regression, Random Forest, LightGBM), a recurrent neural network (GRU), and transformer architectures (ALBERT and FinBERT). Using a normalized corpus of 32,583 financial news items, we implement a standardized experimental pipeline with grid and random search for hyper-parameter optimization under a 60/20/20 data partition. Our evaluation framework moves beyond point estimates by employing nonparametric bootstrapping (n=2,000) to quantify statistical uncertainty and generate 95% confidence intervals for Weighted F1-scores and Macro Area Under the Curve (AUC). Results show that the domain-pretrained FinBERT model achieves superior performance (F1: 0.8705 [0.8621, 0.8785]; Macro AUC: 0.9612 [0.9560, 0.9658]). We apply McNemar's test to verify predictive improvements of domain-specific transformers (p<0.001). 

Keywords

Statistical learning

Sentiment classification

Domain adaptation

Financial text analysis

Transformer models

Model evaluation 

Speaker

Yue Zou, Fu Foundation School of Engineering and Applied Science Columbia University

Co-Author(s)

Yijun Gao, Krieger school of Arts and Sciences, Johns Hopkins University
Zhongyan Wang
Yuchen Cao, Northeastern University
Shuo Xu, Computer Science and Engineering Department, University of California San Diego
Hailiang Wang, Georgia Institute of Technology
wenxi sun