interaction terms are in the model and n 1, then there are no valid tests for interactions
or main effects. However, some of the interactions are known to be 0;
then these interaction terms can be combined to serve as the error term in order to
test the remaining terms in the model. However, it would be a rare situation to run
experiments with n 1, because in most cases the researcher would not know prior
to running the experiment which of the interactions would be 0. Hence, in order to
have valid tests for main effects and interactions, we need n 1.