Analysis of Longitudinal Data with Serial Correlations

RES620 – NIH “Longitudual II”

The ultimate objective of our research is the development of S+LONGIST: a next generation software toolkit for the longitudinal studies. This research will make a fundamental contribution to the conduct of public health studies by developing coherent and mature methodology and software for handling serial correlation in longitudinal data. Considerable research has been devoted towards developing the necessary methodology for applying mixed-effects models and estimating equations models. In spite of this research, longitudinal data analysts often do not fully account for the effects of serial correlation. The aim of the proposed research is to overcome the obstacles and extend the benefits of the research performed to a much wider audience of biomedical analysts and practitioners. To achieve this aim, a framework will be developed based upon three approaches: a state space method, an approximate likelihood approach, and a Quasi-likelihood approach to fit longitudinal data with errors from an exponential family distribution. This methodology will be matured by incorporating diagnostic techniques and will be implemented as an object-oriented software module in the S-Plus language. A comprehensive case study guidebook will be developed involving real problems with serially correlated data.