Methods for estimation of dispersion effects in two-level unreplicated factorial designs are studied. The consequences of non-constant variance are discussed. A natural assumption concerning the form of the covariance of location effects leads to a particular normal model. Some linear combinations of the response variables are constructed in order to make a simple structure for inference. The precision of point estimators of dispersion effects, where one is based on experiments with replicates, are compared. A numerical example is given as an illustration of a test. Finally, estimations in fractional designs are described and discussed.
展开▼