What Size Should a Mobile Ad Be?

We present a causal inference framework for evaluating the impact of advertising treatments. Our framework is computationally efficient by employing a tree structure that specifies the relationship between user characteristics and the corresponding ad treatment. We illustrate the applicability of our proposal on a novel advertising effectiveness study: finding the best ad size on different mobile devices in order to maximize the success rates. The study shows a surprising phenomenon that a larger mobile device does not need a larger ad. In particular, the 300*250 ad size is universally good for all the mobile devices, regardless of the mobile device size.