Comments on the weighted regression approach to missing values

Abstract

Abstract: These comments relate to those methods of dealing with missing values of explanatory variables in regression analysis that first "complete" the data by inserting estimates derived from regressions of explanatory variables on each other and then employ some form of weighted regression. It is argued that the choices of weights in the published methods are not optimal and that improvements are possible. This is verified for a simple case and the difficulties to extending the methodology to general cases are discussed.