A0226
Title: A new robust approach for regression analysis of panel count data with time-varying covariates
Authors: Dayu Sun - Indiana University (United States) [presenting]
Abstract: The focus is on robust inference for panel count data with time-varying covariates, which often occur in real-life applications such as medical studies and reliability experiments. Several mean model-based methods that are robust to within-subject correlation structures have been proposed and widely used. However, the robust mean model-based approach usually requires the mean function to be monotone, which may not be realistic when covariate values fluctuate over time. To address these issues, we propose a robust rate model-based procedure with rigorous theoretical justification. Under the rate model, another challenge is to develop variance estimators because the asymptotic variance usually has no closed forms. To our knowledge, the existing literature with similar problems all resorted to computationally intensive numerical methods that may not be robust. We develop novel computationally efficient robust variance estimators with closed forms based on Expectation-Maximization (EM) algorithm. We rigorously show that the variance estimators are consistent regardless of the underlying distribution assumption. An extensive simulation study is performed to demonstrate the superiority of the proposed approach and a real-life study of sexually transmitted infections is used to demonstrate the applicability of this newly proposed approach.
