An Empirical Analysis of Complementarities Between the Consumption of Goods and Advertisements

The standard paradigm in the empirical literature is to treat consumers as passive recipients of advertising by firms, with the level of ad exposure determined by consumer characteristics and targeting technology. This paradigm ignores the fact that consumers may actively choose their consumption of advertising, thus effectively rendering the level of ad exposure endogenous. Endogenous consumption of advertising is common. Consumers can easily choose to change channels to avoid TV ads, click away from paid online video ads (e.g., TrueView ads on YouTube), or discard direct mail without reading advertised details. Becker and Murphy (1993) recognized this aspect of demand for advertising and argued that advertising should be treated as a good in consumers’ utility functions, thereby effectively creating a role for consumer choice over advertising consumption. They argued that in many cases demand for advertising and demand for products may be linked by complementarities in joint consumption. We leverage access to an unusually rich dataset that links the TV ad consumption behavior of a panel of consumers with their product choice behavior over a long time horizon to measure the co-determination of demand for products and ads. The data suggests an active role for consumer choice of ads, and for complementarities in joint demand. To interpret the patterns in the data, we fit a structural model for both products and advertising consumption that allows for such complementarities. We explain how complementarities are identified. Interpreting the data through the lens of the model enables a precise characterization of the sub-population that consumes advertising conditional on exposure, and thereby facilitates a more precise assessment of the treatment effect of advertising under such endogenous non-compliance. The model also enables assessments of the welfare effects of advertising. To illustrate the value of the model, we compare advertising, price responses and welfare under the model to a setup in which ad consumption is not treated as a choice under the control of consumers. We believe the results have important implications for interpreting ad-effects in empirical work generally, and for the assessment of ad-effectiveness in many market settings.