Approximations of the Generalized Wilks' Distribution

Wilks' lambda and the corresponding Wilks' distribution are well known concepts in testing in multivariate regression models.The topic of this paper is a generalization of the Wilks distribution.This generalized Wilks' distribution is relevant for testing in multivariate regression models with monotone missing data.Just as the (usual) Wilks distribution can be approximated by the x2-distribution of Bartlett (1947), the generalized Wilks' distribution can be approximated by x2-distributions in more or less the same way.We use Box transformations to derive the x2-approximations and compare them by simulation.