Year: 1976
Type: article
Abstract: I PREVIOUS studies of the determinants of household food consumption have typically been based either on cross-sectional or on aggregate time series observations.The crosssectional studies have been characterized (often necessarily) by the assumption that all households in the sample are operating in the same market, making it impossible to study the effect of price variation on the family food budget.Further, such data sets could not be used to distinguish short run versus long run behavior patterns.On the other hand, the studies based on aggregate time series could not shed any light on such important "household-specific" questions as the effect of family composition on food expenditures, or on the economies of scale in food consumption.In the present study we are able to relax many of the important limitations of both the cross-sectional and the aggregate time series studies by basing our analysis on an extensive set of sample observations on households over a number of periods of time.The actual data underlying this study are drawn from the Panel Study of Income Dynamics -a data set collected by the Survey Research Center at the University of Michigan.This panel contains budget and other data on five thousand U.S. households over a period of five years, 1968-1972.1Our model of household behavior distinguishes between various sources of income, family composition and various price effects, and allows for the dynamics of adjustment in household behavior.Further, with regard to functional form, we make use of the procedure suggested by Box and Cox (1964) to determine the appropriate degree of curvature.With respect to estimation procedure, the variancecovariance matrix of disturbances is specified in accordance with the assumed nature of these pooled cross-section and time series observations.In addition, the size of our data allows us to separate carefully the hypothesis searching and the hypothesis testing phases of our study.Our procedure is to divide the entire sample into two independent subsamples, using one as an input in the process of specifying our model and saving the other for testing.The initial subsample, containing approximately one-quarter of the households, is used for the purpose of weeding out potential explanatory variables, determining the functional form of the relationship, and specifying the form of the variance-covariance matrix of the disturbances.On the assumption that all observations are mutually independent, there is no presumption that the households not included in the initial subsample should behave in the same way as the households that are included, unless the regularity in behavior which we want to test actually exists.II In formulating our model, we start with the linear expenditure system proposed by Stone (1954) which can be expressed as follows: / n PiXi -ciPi + bi Y - cjPj ) ( 1) where P price, X = quantity consumed, and Y = current income.All subscripts refer to [ 129 1
DOI: 10.2307/1924018
Authors: J. Benus, J. Kmenta, H. Shapiro
Language: English
Subfield: Economics and Econometrics
Domain: Social Sciences
SDG: Zero hunger
Open Access status: green
OpenAlex ID: W1980727235