Orthogonality in multiway classifications (Q1081258)

We point out the difference between weak orthogonality and strict orthogonality of two factors when all other factors (if any) have been eliminated. Necessary and sufficient conditions for weak orthogonality are obtained, and a simple method of computing the sum of squares for such designs is given. An example of a situation with weak orthogonality is given. It is shown that under some mild conditions a connected equireplicate row-column-treatment design with weak orthogonality for every pair of factors must be of Latin-square type.