Identifiability Analysis Using Data Cloning

Abstract

Lack of identifiability in statistical models may hinder unique inferential conclusions. Therefore, the search for parametric constraints that ensure identifiability is of utmost importance in statistics. However, for complex modeling strategies, even acquiring the knowledge that the model is unidentifiable may prove very difficult. In this paper, we investigate the use of Data Cloning, a modern algorithm for classical inference in latent variable models, as a tool for assessing model identifiability. We discuss its advantages and disadvantages and illustrate its use with a dynamic linear model.