Starting Values for EM Estimation of Latent Class Joint Model

Abstract

Latent class model is one of the important latent variable methods for joint modeling longitudinal and survival data. Latent class joint model can handle underlying heterogeneous population, discover subpopulation structure, and incorporate correlated non normally distributed outcomes. The maximum likelihood estimates of parameters in latent class joint model are generally obtained by the EM algorithm. Finding the starting values is one of the major issues to implement the EM algorithm successfully. In this article, initial value formulas are provided, a simulation study is conducted to show that the proposed starting values perform very well, and two illustrative examples are presented.

Keywords:

• Effective age
• Heterogeneous random effect model
• Longitudinal marker
• Recurrent events

Mathematics Subject Classification:

• 62F10
• 62J05
• 62N02
• 62H12

Acknowledgment

The author would like to thank Dr. Daniel Stram for his helpful suggestion. The author also appreciates the constructive and invaluable comments of the Editor and referees.