701 – Advanced Statistical Methods for Marketing
Causal Models in Estimation of the Advertising ROI
Igor Mandel
Telmar Group
A new approach to causal modeling was recently proposed and published. Its main differences from more traditional counterfactual framework (potential outcomes, structural graphs and equations) are lying in several features: it doesn't require any assumptions of "potentials", working only with existing data; it works successfully with data where covariates are correlated among themselves, but practically not correlated with dependent variable (thus sharply eliminating the multicollinearity problem), it replaces regression-like paradigm to the concept of intrinsic probability, etc. It was actively used in media planning by Telmar and showed very promising results. In this presentation these results are discussed in combination with another challenging question: how orientation of the media planning not just to people, but to people with specific (highest or lowest marketing value) may change the way the ROI is measured and traced? Questions like that are undoubtedly very important from both practical and theoretical points of view.