Augmented Difference-in-Differences

Marketing scientists often estimate causal effects using data from pre/post test/control quasi-experimental settings. We propose a new, easy-to-implement augmented difference-in-differences (ADID) method that complements existing approaches to estimate the average treatment effect on the treated (ATT) from such data. Its advantage over the difference-in-differences method is that it can better handle heterogeneity between treatment and control units and, hence, requires a less stringent causal identification assumption. Its advantages over more flexible approaches like the synthetic control method are that it is easy to implement, provides easy-to-compute confidence intervals, and can be applied to data where the synthetic control and related methods cannot be applied or may not be well suited. Examples are data with short pre- and posttreatment periods or with a large number of treatment and control units. Using analytical proofs, simulations, and nine empirical applications, we document the attractive properties of ADID and provide guidance on what method(s) to use when. With the addition of ADID in their toolkit, marketers are better equipped to address important causal research questions in a wider range of data structures.