Meta-Analysis Methods in Measuring Brand Ad Effectiveness (303185)

Keywords: Meta-analysis, Bayesian estimation, random-effects models, causal modeling, randomized experiments, online ads, online video ads, Google searches

How do you calculate an overall measure of brand ad effectiveness from hundreds of online ad experiments? How do standard meta-analysis methods fare in this context? We describe several meta-analysis methods used and developed at Google to measure overall brand ad effectiveness given data from multiple individual experiments. We provide a brief introduction to standard fixed- and random-effects meta-analysis models and discuss a Bayesian model we developed that accounts for cross-study correlations between exposed and control responses. We present case studies showing the application of these models to measuring ad effectiveness and discuss challenges faced in our application. This talk complements the introductory presentation on "Measuring Brand Ad Effectiveness."