Variance Component Biases of MLE and PQL Methods in Logistics Linear Mixed Model

Restu Arisanti, I Made Sumertajaya,Khairil Anwar Notodiputro,Indahwati

The methods of Generalized Linear Mixed Models (GLMMs) are extended to provide estimates and variance parameters for Logistics Linear Mixed Model (LLMM). LLMM is one of a special case of GLMMs which have the binomial-distributed response. Maximum Likelihood Estimation (MLE) is a method that commonly usedto estimate the fixed effects and random effects of LLMM. The issue of downward bias estimators which is obtained by the MLE method has received considerable critical attention. The main challenge faced by many researches is obtaining the suitable method for GLMMs. Recent evidence suggest that the Penalized Quasi Likelihood (PQL) method can be applied for GLMMs. The aim of this study is comparing the MLE and PQL methods and briefly reviewthe procedure in reducing the biases of variance component estimation in LLMM especially in longitudinal data. It is shown that the estimation of the variance components with MLE are downward biased so that they tend to reject null hypothesis. The biases of the variance components of PQLmethod involve a penalty to obtain less bias comparably than the MLE. Further, both methods are presented to the modelof longitudinal data.

Volume 12 | Issue 2

Pages: 3296-3300

DOI: 10.5373/JARDCS/V12I2/S20201452