סמינר בשיווק

Ron Berman , The Wharton School of the University of Pennsylvania

Recent research analyzing advertising experiments has shown that using a randomized controlled trial (RCT) methodology may result in underpowered experiments even for very profitable campaigns. The common solution applied to this problem is to increase the required sample size or find alternative methodologies to perform the experiments.

We analyze the case of a company that is able to receive a noisy signal about the customer’s responsiveness to advertising. Our findings show that under some conditions, heterogeneity in customer responsiveness to advertising may allow to reanalyze the data using stratification and increase the power of the analysis. The question that remains is whether these conditions actually occur in reality.

By applying the analysis technique to 23 direct marketing experiments, we develop a technique where advertisers can use information from past experiments to improve the power of analysis in future experiments.

We show that in quite a few of these experiments, power can be improved by using our new analysis technique.