a. Let the random (n xl) vector Y represent a random sample from some composite experiment, where Ee = [0], and Eee' = a 21. Suppose the x-matrix has full column rank, but that the first and second columns of x are nearly linearly dependent and, as a 507 result, the determinant of x'x is near zero, equaling

.273 x 10-7• In this case, although fJ = (X'X)-lx'Y is still an unbiased estimator, it is no longer BLUE

(i.e., it loses its "best" property of having the smallest covariance matrix in the linear unbiased class of estimators).