Inference for Structural Breaks in Spatial Models

Abstract

Testing for structural changes in spatial trends constitutes an important issue in many biomedical and geophysical applications. In this paper, a novel statistic based on a discrepancy measure over small blocks is proposed. This measure can be used not only to construct tests for structural breaks, but also to identify the change-boundaries of the breaks. Asymptotic properties and limit distributions of the proposed tests are also established. To derive such asymptotics, a notion of spatial physical dependence is adopted to account for the spatial dependence structure. A bootstrap procedure is applied to the proposed statistic to handle the asymptotic variance of the limit distribution. The method is illustrated by simulations and data analysis.

Resampling Method for Generalized One-per-Stratum Sampling Designs

“…Remark 4. It is worthy pointing out that block bootstrap is commonly used to make statistical inference for dependent data (Lahiri, 2018;Hala et al, 2020;Chan et al, 2022;Zhang et al, 2023). In this paper, our goal is not to propose a new resampling method.…”

“…Lahiri and Zhu (2006) discussed both fixed and stochastic sampling designs, and a block resampling method was theoretically inves-Statistica Sinica: Newly accepted Paper (accepted author-version subject to English editing) tigated. Also see Lahiri (2018), Hala et al (2020), Chan et al (2022) and Zhang et al (2023). Although valid inference can be made through resampling methods, existing works assumed a fixed sampling density function to generate samples, leading to unsatisfactory spatial balance.…”