Alcohol consumption, beverage prices and measurement error

Abstract

Alcohol price data collected by the American Chamber of Commerce Researchers Association (ACCRA) have been widely used in studies of alcohol consumption and related behaviors. A number of problems with these data suggest that they contain substantial measurement error, which biases conventional statistical estimators toward a finding of little or no effect of prices on behavior. We test for measurement error, assess the magnitude of the bias and provide an alternative estimator that is likely to be superior. Method: The study utilizes data on per capita alcohol consumption across U.S. states and the years 1982-1997. State and federal alcohol taxes are used as instrumental variables for prices. Results: Formal tests strongly confirm the hypothesis of measurement error. Instrumental variable estimates of the price elasticity of demand range from -0.53 to -1.24. These estimates are substantially larger in absolute value than ordinary least squares estimates, which sometimes are not significantly different from zero or even positive. Conclusions: The ACCRA price data are substantially contaminated with measurement error, but using state and federal taxes as instrumental variables mitigates the problem.