|
Abstract:
|
We describe ways to measure ad effectiveness for brand advertisements using online surveys. We estimate the causal effect of ads using randomized experiments. We focus on some technical issues that arise with imperfect A/B experiments-corrections for solicitation and response bias in surveys, discrepancies between intended and actual treatment, and comparing treatment group users who took an action with control users who might have acted. We discuss different methods for estimating lift for different slices of the population, to achieve different goals. We use regression, with a particular form of regularization that is particularly suited to this application. We bootstrap to obtain standard errors, and compare bootstrap methods.
|
