On computing generalized least squares and maximum-likelihood estimates of error-components models with incomplete panels and correlated disturbances

Calculation of the inverse of the error variance-covariance matrix is required for both feasible generalized least squares and maximum-likelihood estimation of the regression parameters in the two-way error-components model. Since in many applications this matrix can be quite large, efficient computational methods for inverting the matrix are therefore needed. Incomplete panels complicate calculation of the inverse, and, to date, an efficient method for calculating the inverse has not been provided for the two-way error-components model in which the disturbances are correlated and the panel is incomplete. This note rectifies this shortcoming.