Srivastava, J. N.; McDonald, Lyman - In: Journal of Multivariate Analysis 1 (1971) 1, pp. 118-128
The class of general incomplete multiresponse (randomized block) models, in which it is not necessary for all responses to be measured on every experimental unit, is considered. For p = 2 responses, the optimum model is obtained under a reasonable cost constraint and the determinant criterion....