GQL inferences in linear mixed models with dynamic mean structure

• Main Author: Sun, Bingrui, 1985-
• Other Authors: Memorial University of Newfoundland. Dept. of Mathematics and Statistics
• Format: Thesis
• Language: English
• Published: 2009
• Subjects: Analysis of covariance; Estimation theory; Linear models (Statistics); Newfoundland studies; University of Newfoundland
• Online Access: http://collections.mun.ca/cdm/ref/collection/theses4/id/57212

Thesis (M.Sc.)--Memorial University of Newfoundland, 2009. Mathematics and Statistics Includes bibliographical references (leaves 54) In some panel data studies for continuous data, the expectation of the response variable of an individual (or individual firm) at a given point of time may depend on the covariate history up to the present time. Also, the response at a given point of time may be influenced by an individual random effect. This type of data are usually analyzed by fitting a linear mixed model with dynamic mean structure. When the distribution of the random effects and error components of the model are not known, the likelihood inferences cannot be used any longer. As a possible remedy, there exists some alternative estimation methods such as bias corrected least squares dummy variable (BCLSDV) and instrumental variables based generalized method of moments (IVGMM), which however may produce inefficient estimates. In this thesis, we develop a new GMM as well as a generalized quasi-likelihood (GQL) estimating approach and demonstrate that they perform well in estimating all parameters of the model, the GQL being in general more efficient than the GMM approach.