|
Abstract:
|
In many experiments, standard parametric theory normality assumptions are not met. Brunner et al. (2002, Wiley) have developed a nonparametric approach to the analysis of factorial designs. As the covariance structure of the original data is generally not preserved for the ranked data, these methods require estimating an unstructured covariance matrix. A randomized complete block design with a factorial structure of treatments is a popular choice in agricultural experiments. However, the number of replications is often limited by practical considerations. Estimating a large number of covariance parameters may therefore lead to poor performance. As a parsimonious, yet justifiable approximate model, we propose heterogeneous compound symmetry. In an extensive simulation study, we compare our proposed method to that described by Brunner et al.
|
