Identifiability and Interpretability of Excess Zero and Overdispersed Longitudinal Count Data Models

• In Event: Experimental Design and Hierarchal Linear Models

Abstract

The data on alcohol consumption among adolescents and young adults from the National Longitudinal Survey of Youth (NLSY97), exhibits excess zeros and overdispersion, necessitating the use of hurdle count data models. Our study addresses the critical issue of identifiability and interpretability in hurdle mixed effects models. We assess structural identifiability through rank analysis and examine the Fisher information matrix and profile likelihood. Practical identifiability is evaluated through simulation studies, bootstrap resampling, and sensitivity analysis. To address identifiability and interpretability challenges, we propose several solutions, including model simplification, regularization techniques, and model diagnostics. The study illustrates the application of these methods to handle identifiability issues effectively, providing reliable parameter estimates for meaningful conclusions in longitudinal count modeling.

Authors

• Nivedita Bhaktha, Indian Institute of Technology Kanpur

• Ann A. O'Connell, Rutgers University