Abstract
The problem of constructing designs to minimize the squared covariance or correlation between the estimates of two linear combinations of the parameters of a linear regression model is first considered. When the minimum is non-zero the covariance criterion can be equivalent to the c-optimal criterion. When the minimum is zero it often can be attained by a class of designs. It is then of interest to optimize a standard criterion over the class. Some analytic and algorithmic results are reported.