Predicting Unobservable Values and Estimating Missing Ones. A Coordinate-Free Approach. Part 1.

Abstract : A coordinate-free approach to linear prediction is developed. The results extend to those obtained by Watson (1972) and they are applied here to the estimation of random effects in mixed linear models. The treatment is self-contained and gives a study of when the BLU and LS predictors are identical. In Part II the methods will be applied to the closely related topic of estimating missing values in the analysis variance. (Author)