Identifiability in Generalized Linear Models with Random Effects

Identifiablity is a necessary property for the adequacy of a statistical model‎. ‎When a model is not identifiable‎, ‎no amount of data cannot determine true parameter‎. ‎In this article‎, ‎well-known concept of identifiablity and it’s properties is reviewed‎. ‎Moreover‎, ‎since non-identifiablity problem in linear mixed effects models and generalized linear models with random effects is very common‎, ‎our main focus is on these models‎. ‎On the other hand‎, ‎statistical software‎, ‎after fitting non-identifiable models‎, ‎don’t usually indicate the problem and show invalid outputs‎. ‎Consequently‎, ‎it is useful to have a way to check model identifiability before fitting. In this regard‎, ‎some new theorems to check identifiability in generalized linear models with random effects are presented‎. ‎data from non-identifiable models are simulated and problems with model non-identifiablity are listed for showing advantages of the mentioned theorems.‎