ON USING LINEAR REGRESSIONS IN WELFARE ECONOMICS

This article consists of two parts. The first part shows that the ordi nary least squares regression coefficient is a weighted average of slo pes between adjacent sample points. When applied to a linear regressio n, with income as the independent variable, the regression coefficient depends heavily on the slopes of high-income groups. The weight of th e highest income decile may well exceed that of the other nine deciles . This may be undesirable, especially if the regression is used for we lfare analysis, because the marginal propensities to consume attribute d to the commodities are determined by the high-income groups. The sec ond part of the article proposes alternative estimators, the extended Gini estimators, that enable investigators to control the weighting sc heme and to incorporate their social views into the weighting scheme o f the estimators.