Abstract
Constructing an appropriate statistical model for an analysis can be difficult, particularly when a complex experimental design has been used. This paper describes a procedure that uses the structure of the design (summarized by the ‘layout structure’) in conjunction with the randomization to define the model terms that are included in a mixed model statistical analysis. Randomization objects (which form the ‘restricted layout structure’ and can be visualized using a Hasse diagram) are developed from the layout structure and the randomization(s) to establish these model terms. The use of the procedure enables consistency of models where the scope for analyst's choice of terms is clearly specified and makes developing the appropriate statistical model more accessible. It also highlights the importance of the role of randomization in model specification.
Additional information
Notes on contributors
Simon T. Bate
Dr. Bate is Principal Statistician in Statistical Sciences Europe, GlaxoSmithKline. His email address is [email protected]
Marion J. Chatfield
Mrs. Chatfield is Statistics Manager in Statistical Sciences Europe, GlaxoSmithKline. Her email address is [email protected]