Executive Summary
When examining consumers’ choices among products, some product attributes might be excluded from the analysis because they are difficult to measure. If these omitted attributes are correlated with the measured attributes, standard estimation procedures that assume independent errors are inconsistent. A typical example is product advertising, which is often only partially measured and yet is determined by the firm in conjunction with the pricing of the product. The inherent nonlinearity of consumer choice models makes correction for this endogeneity difficult.
The authors propose a procedure based on control functions. The use of control functions in statistics dates back at least to the work of Heckman (1978) and Hausman (1978) and consists of creating new variables that “control” for the problematic correlation. The authors adapt the procedure to discrete choice models, describe conditions under which the procedure is consistent, and provide several illustrative specifications. They also discuss the special issues that need to be considered when price is the endogenous variable, under both marginal cost pricing (i.e., competitive) and pricing with market power (e.g., monopoly).
The authors apply the procedure to data on households’ choice among television options, including cable and satellite. The quality of programming by the cable franchise is not completely observed. In these data, the unobserved portion of quality is positively correlated with the price of cable service. Without correcting for endogeneity, the aggregate demand elasticity is positive. In contrast, the control function approach estimates price coefficients that are negative for all income groups, implying an aggregate demand elasticity that is negative.
For comparison, the authors also apply the “product-market” approach that Berry, Levinsohn, and Pakes (1995, 2004) developed and find that the two approaches obtain similar results. The authors point out that the control function approach can be applied in situations for which the product-market approach is infeasible or inconsistent (e.g., when observed demand is zero for some products in some markets, such that the corresponding product-market constants cannot be calculated.) The authors describe how the control function approach can be applied with standard statistical packages (e.g., Stata, SAS).
Biography
Amil Petrin is Associate Professor of Economics in the Department of Economics at the University of Minnesota, a member of the Conference on Research in Income and Wealth, and a Research Associate of the National Bureau of Economic Research. He serves as an editor for International Journal of Industrial Organization and previously taught at the University of Chicago. He works on methods of estimating demand, with a particular focus on valuing the introduction of new goods in differentiated product markets. He has closely examined both minivans in the U.S. automobile market and satellite dishes in the U.S. television market. He has developed a method for estimating production functions and the theory that is related changes in plant-level production functions to aggregate changes in total expenditures. His publications have appeared in Econometrica, Journal of Political Economy, and Review of Economic Studies, and he has presented papers at numerous seminars and conferences.
Kenneth Train is an Adjunct Professor of Economics and Public Policy at the University of California, Berkeley, and vice president of National Economic Research Associates. The author of Discrete Choice Methods with Simulation, he studies factors that affect consumers’ choices among products and the impact of changes in product attributes on demand. For seven years, he served as chair of the University of California, Berkeley’s Center for Regulatory Policy and has testified about consumer behavior in court cases and regulatory proceedings. Currently an associate editor for Journal of Choice Modeling, he has served on the editorial boards of several other journals and regularly gives workshops on choice modeling for marketing and business professionals. He has published more than 60 articles and three books and has received numerous awards for his teaching and research.
Journal of Marketing Research, Volume 47, Number 1, February 2010
View Table of Contents