Inferring Individual Preferences and Intra-Household Dynamics with Aggregate Data: An Application to Targeted TV Advertising

While many marketing models claim to incorporate individual-level heterogeneity into estimation and subsequent targeting, they should more appropriately be called household-level models due to the typical lack of data availability for distinct individuals. In this research, we aim to enhance targeting intra-household individuals in situations where individual behavior and consumption are difficult to track and observe. We present a model to describe intra-household consumption at the individual level while estimating the model using only aggregate data. The model takes into account factors that are likely at play when individuals are members of households (beyond and including individual-level heterogeneity) such as preference revision, behavioral interaction, and decision power. Methodologically, the model accomplishes this from aggregated data using a novel, but widely applicable, data augmentation algorithm where we impute, within a Bayesian framework, individual-level choices constrained by the aggregate observed choices. We applied our model to a household TV viewing and ad targeting setting using Nielsen People Meter (NPM) data. Our model estimation and counterfactual analysis show that ignoring intra-household dynamics could lead to significant bias in understanding individuals' behavior and that the proposed model would enable advertisers to improve the efficiency of targeting intra-household individuals significantly yielding a greater ROI.