TI--PC SOFTWARE FOR ANALYZING LONGITUDINAL DATA

点击翻译按钮获取中文摘要

We will produce a PC software package, called Ti, tailored to the analysis and display of longitudinal data, arguably the most common design for biomedical data collection. Repeated measures studies may have from two to thousands of observations per subject, from one to many subjects, and continuous or discrete outcomes. Data from one subject with many, continuous outcomes is the classical time series (ex: EEG and other cortical potential measurements) while data from many subjects with a few continuous outcomes (e.g., blood chemistries) or a few discrete outcomes (e.g. serologic titers) typify repeated measures.... For these, and many other designs, Ti's Phase I graphics will display the trajectory of individual subjects and of group summaries including options for interactive subsetting. For inference, Phase I will include the Hierarchical Linear Model [Bryk & Raudenbush, 1987] a analyses for paired (two time point) data. Phase II will greatly extend Ti to include management of repeated measures data, graphics that include display of and stratification by time-dependent covariates (e.g., running hematocrits), and interactive data transformation. Phase II modeling options will include Markov chains, the generalized estimating equation approach of [Zeger, Liang & Albert, 1988], other quasi-likelihood models [McCullagh & Nelder, 1983] along with a rigorous treatment of missing data [Little & Rubin, 1987]. Ti will have a pleasant, consistent interface for a wide variety of options and so will facilitate analyses that would otherwise be done on an ad hoc basis using several statistical packages.