MISSING VALUES IN MULTIVARIATE STATISTICS II. POINT ESTIMATION IN SIMPLE LINEAR REGRESSION

摘要

We reviewed in [1] the literature dealing with missing observations in regression analysis. In this second paper, we study the probabilistic behavior of several ways to estimate the linear regression function between two random variables. We derive the mean square error of prediction for each method of estimation. Tables are given to characterize in terms of the correlation coefficient those situations where a given method has smaller mean square error than its competitors.nIn Section 1, we specify the model and present some results concerning the classical least squares method. In Sections 2, 3, 4, we discuss the zero order regression method, a modified zero order method, and the first order regression method, respectively (see [1] for a general description of these methods.) Section 5 contains general conclusions and recommendations for the use of these estimation procedures.