IIContributions to ANOVA in one or more dimensions, Parts I and It
- S. N. Roy, R. Gnanadesikan
- Annals of Mathematical Statistics, Vol
It is well known that so f~r the joint distribution of the latent roots associated with normal multivariate analysis of variance has been considerably more difficult to derive if the effective number of variates is greater than the number of components of the linear hypothesis than if it is the other way around. This report offers both on the nUll and the non null hypothesis a simple method of derivation of the distribution for the former case by throwing it back on the distribution for the latter case, and in this tie-up a pivotal role is played by the distribution of the latent roots connected with the testing of the hypothesis of independence between two sets of variates.