A0419
Title: Coordinate testing for general sufficient dimension reduction methods
Authors: Shih-Hao Huang - National Central University (Taiwan) [presenting]
Kerby Shedden - Statistics (United States)
Hsin-wen Chang - Academia Sinica (Taiwan)
Abstract: In modern data analysis, the number of covariates is often large, and the relationship between covariates and response is often complex. Parametric regression risks model misspecification, while nonparametric regression suffers from the curse of dimensionality. Sufficient dimension reduction (SDR) regression provides a flexible alternative, summarizing covariate effects through a few linear combinations without imposing a specific functional form. While SDR methods have been extensively studied, coordinate testing, which assesses the contribution of a set of linear combinations of covariates, has been largely overlooked. To address this gap, a novel method is proposed that transforms the coordinate testing problem into a dimension testing problem by applying appropriate residualization. Since dimension tests are well-established, this method allows practitioners to leverage existing inference tools within the SDR framework.
