A1336
Title: Hasse diagrams and covariates
Authors: Hans-Michael Kaltenbach - ETH Zurich (Switzerland) [presenting]
Damianos Michaelides - Cyprus Institute of Neurology and Genetics (Cyprus)
Simon Bate - GlaxoSmithKline (United Kingdom)
Abstract: Hasse diagrams provide a visual representation of the blocking and treatment structure of an experimental design and the treatment randomization. They allow automated specification of an adequate linear model and are a powerful nontechnical means for discussing design alternatives with non-specialists. In many experiments, additional information in the form of covariates is available and should be used in the analysis. However, covariate adjustments are usually seen as part of the analysis and not the design of an experiment, and hence have not yet been represented in Hasse diagrams. The aim is to present methods for integrating covariates into Hasse diagrams to represent the full ANOVA table, correctly account for covariates' degrees of freedom, and allow automated specification of analysis of covariance models. The use of covariate adjustments is showcased in several standard designs of increasing complexity. It is also shown how this methodology can be extended to the modelling of continuous factors, thus allowing both categorical and continuous factors to be represented on Hasse diagrams.
