An Empirical Analysis of a Two-Sided, Ad-Financed Media Platform with Targeted Advertising

I introduce and estimate a structural empirical model of a two-sided, ad-financed media platform that incorporates targeted advertising. Heterogeneous users decide how much of their time to spend on the platform given the ad levels they face, which is effectively the price paid. Advertisers decide which demographic types of users to target with their advertisements, and their returns to advertising vary across types. The platform selects an ad level for each user demographic type to maximize ad profits. I use data on Facebook usage and advertising to estimate the model. Because no public source of these data exists, I create a novel dataset derived from a survey I conducted of US Facebook users (n = 2,287). Descriptively, I find wide heterogeneity in Facebook usage across users, with the bulk of users spending less than 20 minutes per day on the platform but a long right tail of users spending over an hour. I find that on average users encounter 3.4 ads per minute. This figure varies widely across users (standard deviation of 3.0), which distinguishes Facebook from traditional media. Using these data, I find evidence of an ad nuisance effect on Facebook, but that usage is inelastic with respect to ad levels on average. I estimate mean returns of advertising to different demographic types and find that they vary quite widely. I show that mean returns are influenced by differences in ad responsiveness by different types (for instance, differences in average click-through rates) but not influenced by differences in usage as would be expected of advertiser demand in traditional media outlets. I use the model estimates to conduct a counterfactual analysis in which Facebook is no longer permitted to target ads across demographic types. I find that when underlying mean returns to advertising do not fall, Facebook can increase ad levels to completely offset any potential loss in revenue at the expense of ad-averse users. While the predicted average ad price falls, advertisers are worse off overall from the loss of targeting. When underlying mean returns to advertising decrease due to the lost ability to target, Facebook profits fall, advertiser surplus falls, and ad levels increase which harms users.