Model Comparison for Dependent Generalized Linear Model
Year: 2016
Type: preprint
Abstract: The classical Bayesian information criterion (BIC) is derived through the stochastic expansion of marginal likelihood function under suitable regularity condition when models are correctly specified. However, despite of its popularity, mathematical validity of BIC for possibly misspecified models with complicated dependence structure is often ignored. Thus it is important to extend the reach of the classical BIC with rigorous theoretical foundation with allowing model misspecification and asymptotic mixed normality of estimator. In this paper, we will prove the stochastic expansion of marginal quasi-likelihood function associated with a class of possibly misspecified generalized linear models for dependent data.
Source: arXiv (Cornell University)
Author: Shoichi Eguchi
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2222811565