Distributions and Transformations for Family Expenditures

Stuart Scott and Daniel J. Rope

Abstract

Family expenditure distributions from the U.S. Consumer Expenditure Survey, at both detailed and aggregate levels of spending categories, are highly skewed. Lognormal and gamma-family distributions, including the generalized gamma distribution, are fitted and compared for selected expenditure items. In addition, Box-Cox transformations to achieve approximate normality are carried out. To select the transformation, direct methods and methods based on the models (cf. Hernandez and Johnson, 1980) are applied. These expenditure data are utilized in numerous analytic studies. Implications for data users are discussed.