Abstract:
quite a lot of methodology has been developed for the analysis of longitudinal studies, stemming from clinical trials, epidemiology, and other studies in humans. For example, hierarchical models are becoming ever more frequently. Such hierarchical models are standard in the analysis of longitudinal data, too to account for the correlation steaming from the repeated measures nature. This study will be dedicated to model models for longitudinal continuous, firmly rooted in hierarchical models such as the linear mixed model. One finds, coupled with methodological development, also the availability of standard software tools, including SAS, Stata, SPlus, R, etc. The Bayesian implementation of the models will also be explored using the freely available software WinBugs. The two approaches will then be applied on data set from the Jimma Infants longitudinal growth study. The result demonstrated that the ML estimates of the random-effects standard deviations are smaller than the corresponding REML estimates which is different result from the Bayesian. The estimated within group residual standard deviations are identical. In general, the fixed-effects estimates obtained using ML, REML and Bayesian techniques are almost similar. The mean evolution of the upper arm circumference of infant for boys and girls is not different. For infants given supplementary food and without supplementary food their mean evolution is not different. The linear mixed effect model estimate of the fixed effect obtained using likelihood and Bayesian techniques are almost similar but with different random effect standard deviations.
