Causal effects of binary, continuous and multivariate treatments using propensity score methods in a marketing context

In this thesis we aim to estimate the causal effect of multiple advertisements on the tune-in of a TV program. We contribute to the current literature by applying propensity score methods in a generalized problem setting and relatively new domain. Furthermore, we introduce a combination of average and treated dose-response functions, investigate estimating treatment effects for a low dimensional multivariate (instead of bivariate) treatment variable and introduce a smooth coefficient model for multivariate treatments. In particular, we study the effect of multiple TV advertisements on the tune-in of the season premiere of America’s Got Talent in 2016. Starting from the most commonly studied case in the literature, binary treatments, we extend the analysis to continuous and multivariate treatments. We explore the use of many different methods and prefer CBPS as treatment assignment model for binary treatments and Poisson regression for continuous/multivariate treatments in our case. We find small treatment effects, depending on the treatment variable(s) used. Additional exposures to advertising have a positive impact on the probability of tune-in. Furthermore, the results suggest more recent advertisements have a higher impact and advertisements on the same channel (NBC) are most effective.