Generalizable and Robust TV Ad Effects
41 Pages Posted: 16 Nov 2018
Date Written: October 26, 2018
Much of the empirical literature exploring the economics of advertising may not be generalizable because it uses a case-study model of research, finding a particular effect in a single category and exploring the implications of that effect only in that category. Publication bias may further distort our understanding of the distribution of realized advertising effects if it discourages researchers from pursuing projects where a null effect may exist. Additionally, empirical identification in many studies of advertising can suffer due to a lack of exogenous variation. In this paper, we study the effects of TV advertising across a broad range of brands and categories, which allows us to characterize the full distribution of advertising elasticities. We also evaluate the sensitivity of our results to different identifying assumptions that are frequently employed in the literature. Our distributional analysis provides insights into i) whether or not effects found in the literature are generalizable, ii) the extent to which null effects may be present, and iii) the need to carefully address identification in any empirical study of advertising effectiveness. We find that censoring results that are non-significant or negative biases upward the mean and median of the own advertising elasticity distribution by a factor of about five. This result is robust to various identifying assumptions.
Keywords: Advertising, Publication Bias, Generalizability
JEL Classification: L00, L15, L81, M31, M37, B41, C55, C52, C81, C18
Suggested Citation: Suggested Citation