Nonlinear Split-Plot Design Model in Parameters Estimation using Estimated Generalized Least Square - Maximum Likelihood Estimation

Abstract

This research aimed to provide a theoretical framework for intrinsically nonlinear models with two additive error terms. To achieve this, an iterative Gauss-Newton via Taylor Series expansion procedures for Estimated Generalized Least Square (EGLS) technique was adopted. This technique was applied in estimating the parameters of an intrinsically nonlinear split-plot design model where the variance components were unknown. The unknown variance components were estimated via Maximum Likelihood Estimation (MLE) method. To achieve the numerical stability in the iterative process of estimating the parameters, Householder QR decomposition was used. The results show that EGLS method presented in this research is available and applicable to estimate linear fixed, random, and mixed-effect models. However, in practical situations, the functional form of the mean in the model is often nonlinear due to the dynamics involved in the system process.