الفهرس | Only 14 pages are availabe for public view |
Abstract Linear modelshave been the dominant approach for the analysis of longitudinal data when the outcomeis continuous, in many applications the pattern of change is more faithfully characterizedby a function that is non-linear in the parameters.longitudinal studies are alsomore prone to problems of missing data and attrition. This thesis consists of five chapters: In Chapter 1,a brief overview of the ideas underlying longitudinaldata is give.The aim of Chapter2is to developthe linear mixed effects model motivated by two-stage random effects formulation of the model and the two -stage analysis Chapter 3 discusses the problems raised by missingvalues, which frequently arise in longitudinal studies and the method for handling missing values.Chapter 4 is devoted to presentmore details concerning maximum likelihood estimation of β and θ, the main idea behind REML estimation is devoted to separate that part of the data used for estimation of from that used for estimation of β. Chapter 5 contains all the new results obtained in this thesis the main concern of this chapter the sensitivity analysis of longitudinal data with intermittentmissingvalues.we developed non-normallongitudinal data models. The concentration is mainly on the missing values when the pattern isintermittent and the mechanism is nonrandom, in which random or completely random missingdata be considered as special cases. This enables us to assess the impact of the distributionalassumptions on the underlying parameters. |