Two-way selection of covariables in multivariate growth curve models Original Research Article

The Growth Curve model introduced by Potthoff and Roy [1] has provided a general format for a variety of growth and repeated measures studies. Statistical inference of this model has often been based on the analysis of covariance model (see e.g. [2], wherep measurements are partitioned into theq measurements of the main variables and onp − q covariables. Under the general unstructured model for covariance choosing the full set ofp − q covariables results the maximum likelihood estimates (ML) of the model parameters. However, in many practical situations a more efficient estimator can be obtained by choosing fewer covariables. In this paper we propose a computationally efficient method for choosing covariables. This procedure, which is called the two-way selection, is based on the efficiency considerations and on an ordinary variable selection procedure. The method is compared to the method proposed by Fujikoshi and Rao [3].