Abstract
Consider a linear model with targeted parameters . We derive a necessary and sufficient condition on the matrix M for the average variance of functions
to be proportional to the A value for comparing designs. This establishes the full range of interpretations of the A criterion.