Evaluating the Performance of Weight Functions with No Constraints on Function Types

Abstract

Listen

Translate article

Abstract In weighted linear regression, the variance of the error terms is estimated by a weight function. Methods for testing two competing weight functions exist provided one of the functions is a special case of another weight function. A method based on an index of fit and bootstrap sampling, which allows two unrelated weight functions to be compared, is given. This method is tested on six artificial data sets of different sizes. For data sets of size 100 and larger, the proposed method could detect the difference between the weight functions σ²f1(xi) = σ²e1xi and σ²f2(xi) = σ²xi2 at the 95% level. For. Sci. 40(4):787-793.