Abstract A parametric regression model may comprise several correlated individual regression equations, and these equations may have common and different unknown parameters. In such a situation, the common unknown parameter vectors in these equations can be estimated individually or simultaneously according to various available statistical inference methods. The purpose of this paper is to provide an integral account of two classic objects in regression theory: the best linear unbiased predictors (BLUPs) and the best linear unbiased estimators (BLUEs) on common and different unknown parameters in two correlated linear models with common and different parameters. We first introduce some reduced models associated with the two correlated linear models. We then define and characterize predictability and estimability of all unknown parameter vectors in the two correlated models and their reduced models, and derive analytical formulas for calculating the BLUPs and BLUEs of all unknown parameter vectors in these models by means of a constrained quadratic matrix optimization method. We also discuss a variety of theoretical properties of the predictors and estimators.
Read full abstract