B1257
Title: Cross-sectional versus longitudinal design for repeated measures: A comparison
Authors: Rainer Schwabe - Otto-von-Guericke University Magdeburg (Germany) [presenting]
Abstract: In experimental situations there is often a substantial variability of the experimental units. This typically arises in bio-sciences but may also appear in engineering experiments caused by varying quality conditions of the material. Since experimental units generally stem from a larger entity, it is commonly assumed that the impact of the experimental units is properly described by random coefficients. Then the resulting observations will be correlated within each experimental unit when repeated measurements are taken. If the random effects are associated with the experimental conditions, then they cause additionally heteroscedasticity of the single observations which has strong impact on the performance of an experimental design. Optimal designs are derived in a longitudinal setup, where experimental conditions may vary within observational units and are the same for all units, and in a cross-sectional setup, where the experimental settings remain fixed within each observational unit but may vary across units. The optimal designs obtained can have quite different structures, and their relative efficiencies can become quite low. Therefore the experimental setup has to be chosen properly to fit the experimental situation.
