| Summary: | 碩士 === 國立東華大學 === 應用數學系 === 92 === In many follow-up studies of medicine and sociology, repeated observations of the outcome variable along with several covariate variables are taken at unequally
spaced time points. When the main interest of the study centers on the pattern of change in the outcome over time, it is common to employ Markov regression models in continuous time. The parameters in these models are often estimated by the maximum likelihood method with certain
iterative procedures. Previous estimating procedures are, however, often limited to either homogenous models for which the individuals are followed up at the same points, or
piecewise homogenous models for which the outcome and covariate variables are observed at the same time. In contrast, the estimating procedure of Chao (1996) can be applied to piecewise homogenous models for which the outcome and covariate variables need not be observed at the same time. In this thesis, Chao''s procedure applied to the log-linear transitional model is implemented with a Fortran 90 program. In addition to parameter estimation, certain diagnostic tools such as summary residuals and goodness of fit measures are also incorporated into the program for model checking. To illustrate the utility of the program, we will analyze a longitudinal data set on blood pressure obtained from a community-based follow-up study of cardiovascular diseases.
|
|---|
