Use DiD for staggered treatment adoption

Imagine you are carrying out a phased release across 50 regions spanning 2025-06-01 to 2025-08-15, where the metric of interest is weekly revenue per user. Put together a difference-in-differences design that stays valid under non-uniform treatment effects. Lay out: (a) when two-way fixed effects (TWFE) regression becomes biased and what estimand it actually recovers; (b) how to apply Callaway–Sant’Anna or Sun–Abraham, covering group-time average treatment effects and aggregation weights; (c) an event-study arrangement with an appropriate baseline period and the way you would plot it; (d) pre-trend checks using a joint F-test and what to do if they fail; (e) inference when only a small number of clusters are available, using wild cluster bootstrap and accounting for serial correlation; (f) how you would bring the DiD results together with a matched propensity-score analysis.

Overview: The item assesses skill in panel-data causal inference, in particular difference-in-differences with staggered rollout timing, varying and time-varying treatment effects, event-study specification, pre-trend diagnosis, cluster-robust inference, and comparing with propensity-score matching.