LWDID: Stata module to implmenent Rolling Difference-in-Differences Estimator for Small-N and Large-N Panel Data

lwdid implements the transformation-based rolling Difference-in-Differences estimators developed in Lee and Wooldridge (2025, 2026a). The command provides a unified implementation for panel data settings with either a large-N or a small-N cross-sectional dimension, allowing treatment effects to be estimated under both the staggered treatment adoption and the common timing case. The central idea is to transform outcomes within each unit to remove pre-treatment means, trends, or seasonal components, yielding residualized outcomes. These transformed outcomes allow treatment effects to be estimated using simple cross-sectional regressions in each post-treatment period, facilitating both overall and period-specific ATT estimation. By default, lwdid uses the large-N procedure of Lee and Wooldridge (2025), which is designed for panels with a large cross-sectional dimension and allows for heterogeneous treatment effects and unit-specific heterogeneous linear trends. When the cross-sectional dimension is small (small-N), conventional large-N inference may be unreliable. In such settings, specifying the small option invokes the exact small-sample inference procedures developed in Lee and Wooldridge (2026a). The seasonal-adjustment options demeanq, detrendq, demeanm, and detrendm are currently available only under the small-N implementation. Based on the treatment cohort variable specified in gvar(), lwdid automatically detects whether the design involves a single treatment cohort (common timing) or multiple cohorts (staggered adoption) and applies the appropriate estimation procedure.