CANSSI CAST Speaker:
Dr. Cindy Feng
Associate Professor, Department of Community Health and Epidemiology
Faculty of Medicine
Dalhousie University
Title: Residual Diagnostics for Count and Censored Regression Models
Abstract: Diagnosing regression models is essential for evaluating model fit and detecting discrepancies between the model and data. While traditional residuals like Pearson and deviance residuals are effective for linear models, they face limitations when applied to count and censored regression models. Count data often produce residuals that deviate significantly from normality, and censored data introduces additional complexities that traditional methods struggle to address. In this presentation, we explore the challenges and innovations in residual diagnostics for count and censored regression models. We evaluate randomized quantile residuals (RQRs) and their performance in diagnosing count regression models, including zero-inflated models. Despite their established use, RQRs have not been extensively studied in this context, and our research addresses this gap through simulation studies.
Additionally, we introduce a novel approach for censored regression models using normalized randomized survival probabilities (RSPs). This method transforms the survival probabilities of censored observations into uniformly distributed random numbers, which are then converted into normally distributed residuals. Our study demonstrates that these normalized RSP residuals provide a robust diagnostic tool for censored regression models, effectively detecting model misspecification that traditional residuals might miss. Our findings highlight the effectiveness of these novel residual diagnostics in addressing the unique challenges posed by count and censored regression models. These advancements offer valuable tools for improving model assessment and diagnostic accuracy in statistical analysis.
Day & Time: Friday, September 13, 2024 at 3:00pm
Location: LT 9-118 Zoom (online)*
For more information, visit CANSSI Ontario.
*Counts toward seminar attendance for MSc and PhD students in Math & Stats. This seminar will be connected virtually in LT 9-118 so in-person attendance is required.