442 – Contributed Oral Poster Presentations: Section on Statistical Consulting
Rank-Based Statistical Methods for Longitudinal Studies
Raymond Hoffmann
Medical College of Wisconsin
Longitudinal studies often have distributions that can not be described by a simple generalized linear model. When this occurs an alternative distribution is needed to model the data. In particular, a rank based longitudinal method with spearman rank statistics to estimate the correlation structure will provide a more general alternative solution to this problem. It also provides a more general solution to the shape of the response curve, which might otherwise not allow comparisons between groups.
The bootstrap will be used to obtain the estimates of the functional and correlational parameters in comparison with the corresponding Wald statistics. The results will be base on simulated data that (1) can be fit by a GEE model and (2) simulated data that cannot be described by a GEE because of the shape of the response distribution.