Determinants of Fast Food Consumption

Socioeconomic determinants are investigated for both the likelihood of consuming fastfood and household expenditure on fastfood using the 1994-98 USDA Continuing Survey of Food Intakes by Individuals. The logit model is used to estimate an empirical relationship between probability an individual will consume fastfood and socioeconomic variables. The Tobit model is used to estimate an empirical relationship between expenditure on fastfood and socioeconomic variables. Significant socioeconomic variables impacting the likelihood of consuming fastfood and household expenditure on fastfood included age, income, household size, hours at work, eating occasion and education level. "Determinants of Fast Food Consumption" Page 3 of 27 INTRODUCTION Food away from home is a large and growing component of U.S. food expenditure. Eating and drinking purchases have dominated food away from home expenditures over the last three decades (Putnam and Allshouse). As a share of disposable personal income, food away from home expenditure has risen from 3.6% in 1970 to 4.1% in 1997, while food at home has decreased from 10.2% in 1970 to 6.6% in 1997 (Putnam and Allshouse). Given this change in food consumption, there is a considerable need to identify the determinants of fast food consumption and the implications of these changes for the fast food industry. Jekanowski, Binkley, and Eales examined the effect of price, income, and demographic characteristics on fast food. They suggest that growth in the fast food consumption was related to an increasing supply of convenience, their research utilized aggregated fast food measures at the Metropolitan Statistical Area level, using a composite price index for fast food price. It seems consumers would not increase consumption of fast food unless they had incentive to do so, in other words fast food must be relatively cheap with respect to price, time, and/or taste. Ekelund and Watson also found that fastfood consumption was empirically related to opportunity costs of the household. They argue that preparing ethnic foods has a relatively high time cost and that the effect of this cost is shown in the decreasing proportion of ethnic to total fastfood sales. McCracken and Brandt identified household income, time value, size, and composition as important determinants of total household expenditures on food a way from home. They found that the importance of these factors varied between conventional restaurants, fastfood facilities, and other commercial establishments. Decomposition of Tobit elasticities indicated the importance of household size, income, and time value on market participation. "Determinants of Fast Food Consumption" Page 4 of 27 The main objective of our paper is to identify and quantify the relationship between an individual's socioeconomic characteristics and their decision to consume fast food and their household expenditure decision. The paper proceeds in the following manner. First, data are discussed. Second, empirical issues are addressed. Third, results are presented and interpreted. Finally, conclusions are provided. DATA The data set used is the 1994-98 USDA Continuing Survey of Food Intakes by Individuals. The USDA survey is designed to gather data on food intake by individuals to assess nutritional intake and pesticide residue exposure from food consumption. The data set also contains demographic characteristics of survey respondents as well as the source of the food item. Possible sources included fast food restaurants, which were separable from other food away from home. Thus, observations of food consumption included socioeconomic or demographic variables and whether or not the food item was purchased from a fast food restaurant. These data are adequate for estimating the probability of individuals consuming fast food. Individual expenditure data were not included in the USDA individual consumption survey. However, household expenditure on fast food was elicited in the household survey. Hence, individual and household surveys were linked by identification number allowing demographic variables of the head of household to be applied to the corresponding household. Household survey participants were allowed to enter household fastfood expenditure estimates by week or month. The survey fastfood expenditure responses were converted to annual expenditure estimates for the purposes of this study. "Determinants of Fast Food Consumption" Page 5 of 27 Table 1 provides descriptive statistics on selected variables from the individual and household data as well as a description of each variable. There were 217,939 individual consumption observations. The data indicated that the average individual was 22 years of age and worked about 10 hours per week. Observations of age ranged from individuals less than 1 year of age to 90 years of age. Individuals in the survey indicated they spent as much as 140 hours at work. The largest segment of individuals were college educated and lived in suburban areas. When head of household demographics were projected onto the household there were only 7992 household surveys with a corresponding head of household individual survey. The average age of the head of household was 51 years of age. The average household income was $40,123 in the data set. Only 1.7% of food consumption decisions in the data were related to fastfood consumption. Furthermore, only 57% of food consumption was for main meals, breakfast, brunch, lunch, dinner, or supper, leaving 43% of food consumption that was non-meal associated. Food consumption was evenly spread throughout the day as dinner or supper comprised of 21% of eating occasions, while breakfast and brunch or lunch were both 18% of eating occasions. "Determinants of Fast Food Consumption" Page 6 of 27 EMPIRICAL ISSUES Consider a relationship between fastfood, Y, and a matrix of socio-demographic factors and income, X. Formulating a linear regression model yields Y X e = β+ where e represents unknown error terms and β are parameters to be estimated. Given the censored survey data for fastfood, two limited dependent variable models were estimated. A logit model is specified to identify the relationship between sociodemographic variables and an individual's likelihood of purchasing fast food. The probability of purchasing fast food or “Participation Decision” is modeled by ) 1 /( ) 0 ( ) 1 ( i i BX BX i e e y P + = > , where household variables include the number of individuals in the household, the geographical region the household is located within, whether the household is located in an urban or rural area, and household income. Individual Demographics include age, typical number of hours spent working per week, typical number of hours spent watching television per day, sex, race or ethnic origin, and the highest education level achieved. Other variables not really household or individual related include the month and day of the week the food item was consumed, the name of the meal (i.e. breakfast, lunch, etc.), and whether or not the food item was purchased at a fast food restaurant or not. The “Expenditure Decision” is modeled at the household level with household expenditure as a function of sociodemographic variables. For the Tobit regression model (Tobin 1958), household fast food expenditure is modeled as )] / ( / ) / ( )[ / ( ) ( ) 2 ( σ β σ β φ β σ β i i i i i X X X X y E Φ + Φ = , "Determinants of Fast Food Consumption" Page 7 of 27 where ( ) and ( ) φ Φ represent the normal probability density and cumulative distribution functions. Head of Household variables include sex, age, hours typically spent at work, and race or ethnic origin. Household variables include number of individuals living in the household, household income, geographical region, and whether the household is in an urban or rural area. To complete the model specification of the independent variables the multivariate polynomial functional form was assumed. Based on the literature review and preliminary investigations of the data, a set of explanatory variables were chosen and it was hypothesized that a second order multivariate polynomial maintained sufficient flexibility to capture responses from the explanatory variables. First and second order interaction terms were included to test for significance of the relative influence of each variable on one another. The model specification the structure of Xβ is defined as: (3) ∑ ∑ ∑ ∑ ∑ ∑ > = > = > = > + + + + + + + = z