Department of Mathematical Sciences
Xiaoyu Chen
Monday, April 6th, 2026
9:00AM-11:00AM
Olin Hall 223
Zoom Link: https://wpi.zoom.us/j/4249858010
Speaker: Xiaoyu Chen
Title: Statistical Modeling for Structured Spatial Data: Hierarchical Spatio-Temporal Methods and Socioeconomic Application
Abstract: This dissertation proposal consists of two related research areas in statistical modeling and inference for structured data.
The first area, developed in Chapters 1-3, focuses on statistical modeling and inference for structured spatial data with hierarchical spatio-temporal dependence. We propose a hierarchical spatio-temporal dynamic panel data model that decomposes the response into between-group and within-group components, allowing spatial dependence to operate at multiple levels. To enable scalable estimation for large datasets, the within-group spatial weight matrix is parameterized through a block-circulant structure, yielding a parsimonious representation and substantial computational gains. Under this framework, quasi-maximum likelihood estimators are developed and their large-sample properties, including consistency and asymptotic normality, are established. The proposed methodology will be evaluated through Monte Carlo simulations and applied to U.S. county-level population growth data. In addition, we explore several theoretical extensions, including the incorporation of fixed effects, relaxation of distributional assumptions, and bias-reduction procedures. This research aims to provide a flexible, computationally efficient, and theoretically grounded framework for analyzing large hierarchical spatial datasets.
The second area, developed in Chapter 4, studies the micro-determinants of fertility using the Absolute and Relative Treatment Effects (AbRelaTEs) model. Motivated by the substitution effect of housing appreciation on childbearing decisions, this project investigates how rising housing prices influence fertility intentions and demographic trajectories. The AbRelaTEs framework simultaneously estimates absolute and relative treatment effects, with the logistic model arising as a special case. We will examine the theoretical and empirical advantages of this framework relative to conventional binary choice models, particularly in the presence of heterogeneous and relative treatment effects. Monte Carlo simulations and empirical analysis will be conducted to assess the model’s performance and to uncover heterogeneity across fertility margins. This research aims to demonstrate the importance of using generalized choice models such as AbRelaTEs for reliable demographic and policy inference.
Overall, this dissertation seeks to contribute to both methodological statistics and applied quantitative research by developing new tools for structured spatial data and applying modern statistical models to important socioeconomic questions.
Committee members: Prof. Fangfang Wang (Advisor), Prof. Adam Sales, Prof. Buddika Peiris, Prof. Zheyang Wu, Prof. Zhengjun Zhang (UW-Madison)