B1511
Title: Semi-parametric local variable selection under misspecification
Authors: Michele Guindani - University of California Los Angeles (United States) [presenting]
Abstract: Local variable selection aims to discover localized effects by assessing the impact of covariates on outcomes within specific regions defined by other covariates. The challenges of local variable selection are outlined in the presence of non-linear relationships and model misspecification. Specifically, a potential drawback of commonly used semi-parametric methods is highlighted: even slight model misspecification can result in a high rate of false positives. To address these shortcomings, a methodology based on orthogonal splines is proposed that achieves consistent local variable selection in high-dimensional scenarios. The approach offers simplicity, handles continuous and discrete covariates, accommodates multivariate covariates, and provides theory for high-dimensional covariates and model misspecification. Settings with either independent or dependent data are discussed. The proposed approach allows for the inclusion of adjustment covariates, enhancing flexibility in modelling complex scenarios. Simulation studies illustrate its application with independent and correlated data and two real datasets. One dataset evaluates salary gaps associated with discrimination factors at different ages, while the other examines the effects of covariates on brain activation over time.
