We show how to use micro-level survey data from a tracking study on brand awareness in conjunction with data on sales and advertising expenditures to improve the specification, estimation, and interpretation of aggregate discrete choice models of demand. In a departure from the commonly made full information assumption, we incorporate limited information in the form of choice sets to reflect the fact that consumers may not be aware of all available brands at purchase time. We find that both the estimated brand constants and the price coefficient are biased downward when consumer heterogeneity in choice sets is ignored. These biased estimates lead to costly mistakes in firms price setting. In addition, the tracking data allow us to identify separately two processes by which advertising influences market shares. We find that advertising has a direct effect on brand awareness (inclusion in choice set) in addition to its effect on consumer preferences (increase in utility). This improved understanding of how advertising works enhances our ability to make policy recommendations.
