A Continuous-Time Mover-Stayer Model for Categorical Longitudinal Data

• Main Authors: Yi-Ran Lin, 林逸然
• Other Authors: Wei-Hsiung Chao
• Format: Others
• Language: en_US
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/g8fxd4

碩士 === 國立東華大學 === 應用數學系 === 95 === Quite a few Markov regression models have been proposed to study the pattern of change in a categorical response over time in the panel data setting. An implicit assumption of the Markov model is that all individuals are subject to specific transitions with positive probabilities if the corresponding transition intensities are not structural zeros. This assumption is not appropriate in certain medical studies when only a fraction of the population is subsceptible to get a disease or make specific transitions between disease stages. Following the work of Cook, Kalbfleisch and Yi (2002), we propose a generalized mover-stayer model for conditionally piecewise time-homogeneou Markov process that is more appropriate in practice. The mover-stayer indicators are assumed to be multinomially distributed which is useful when all states compete to be a stayer state for an individual. A Fisher scoring algorithm is described which facilitates maximum likelihood estimation based on the first derivatives of transition probability matrices. We discuss the performance of the algorithm in a real data analysis as well as possible improvements for future research.