Primary Submission Category: Matching, Weighting
Matching Methods for Difference-in-Differences with Multiple Time Periods: Evaluating the Equality of ATT Estimates Across Time
Authors: Junho Jang, Yitae Kwon, Kwonsang Lee,
Presenting Author: Junho Jang*
In observational studies with multiple time points, testing for homogeneous causal effects, such as the Average Treatment effect on the Treated (ATT), is crucial for evaluating treatment efficacy over time. This paper introduces a novel testing framework that accommodates arbitrary combinations of treatment initiation and post-treatment time points. For estimation, difference-in-differences (DID) is combined with matching to focus on post-treatment periods for treated units. Our testing framework involves two main steps. First, a confidence set for the common treatment effect is constructed to narrow the range of plausible parameters. Second, a randomization-based test is conducted within this confidence set to assess the equality of treatment effects. This approach extends multivariate location testing to partially matched sets. Furthermore, the relationship between matched set structure and test power is theoretically explored, providing insights to guide matching design in practice. To illustrate its application, we use this framework on Health and Retirement Study (HRS) data, testing and summarizing treatment equality across time periods.