Determinants of Household Expenditures on Alcohol

This paper examines the determinants of household alcohol expenditures by using a nonnormal and heteroscedastic double-hurdle model to accommodate zero observations in the sample. The model is a generalization of the double-hurdle model estimated in previous studies of alcohol consumption. We also examine the effects of explanatory variables by calculating and decomposing the elasticities. Findings support the use of a more generalized error distribution. Income, region, education, and household demographics are among the significant determinants of alcohol expenditures. DETERMINANTS OF HOUSEHOLD EXPENDITURES ON ALCOHOL There is long-standing interest in the economic aspects of alcoholic beverages, although formal economic models of alcohol consumption are a more recent phenomenon (Hilton 1993). Findings on the price responses of alcohol demand have obvious policy implications!in promoting tax revenues and controlling consumption through taxes (Ornstein 1980). The identification of non-price determinants of alcohol consumption is also important because such information often plays a crucial role in the prevention of alcohol abuse and alcohol-related problems (Hilton 1993). There have been numerous studies on the demand for alcoholic beverages. Ornstein (1980) and Ornstein and Levy (1983) provide surveys of the earlier studies, and Leung and Phelps (1993) review the studies that have appeared during the last decade. In addition, Selvanathan (1991) reviews the recent literature on alcohol demand systems. Most of the earlier studies have been based on aggregate time series. However, a notable trend among the more recent studies on alcohol consumption and expenditures is the growing use of survey data. The examples include Atkinson, Gomulka, and Stern (1990), Blaylock and Blisard (1993), Heien and Pompelli (1989), Pompelli and Heien (1991), and Yen (1994). The use of microdata allows examination of the effects of detailed household characteristics and provides the degrees of freedom to estimate a large number of parameters. However, microdata also present a unique problem: the sample often contains a significant proportion of observations with reported zero expenditures. Standard econometric techniques not accounting for this data feature lead to biased and inconsistent parameter estimates (Maddala 1983). Researchers have often used the Tobit model (Tobin 1958) to estimate demand relationships with limited dependent variables. However, parameterization of the Tobit model is known to be very restrictive, which makes it unpalatable for empirical analysis. Recent analysts have used the double-hurdle model in modeling alcohol demand (Blaylock and Blisard 1993; Yen 1994). The double-hurdle model features two separate stochastic processes for participation and consumption; it allows for examining the determinants of both participation and consumption decisions and provides more useful insights on consumer behavior than the Tobit model. Applications of the double-hurdle model in other areas of food demand include Haines, Guilkey, and Popkin (1988), Reynolds (1990), and Wang and Jensen (1994). The model specification and estimation procedure of the double-hurdle model require specification of the distribution for the error terms of the participation and consumption equations. At issue are a number of aspects related to the error distribution: (a) dependence between the error terms, (b) heteroscedasticity, and (c) shape of the distribution. In previous studies of alcohol consumption, the error