Computation of variance components by the MINQUE method (Q1069247)

We present a new method of computing C. R. Rao's MINQUE in variance component models \((y=X\beta +U_ 1\xi_ 1+\cdots +U_ p\xi_ p)\), which requires only inversion and storage of \(n_ i\times n_ i\) matrices, where \(n_ i\) is the number of columns in \(U_ i\). In many cases most of these matrices are of diagonal form. In particular, the derivation of MINQUE equations for univariate nested classification models does not need any inversion of matrices.