Evaluate Miami Ultrafast impact on orders

Two months ago, Instacart rolled out an Ultrafast delivery option in Miami. You have 52 weeks of weekly order counts for each geography; Miami is considered treated starting at week $t_0$, while the other geographies have not yet launched. Assess the causal effect on user orders. Address:

(1) Control selection: suggest two approaches—for example, matched markets built from propensity scores on pre-period covariates and outcomes, or a synthetic control that minimizes pre-period RMSE—and explain how you would validate whichever control you choose.

(2) Assumptions: state the parallel-trends assumption; describe how you would set up and read pre-trend tests and an event-study plot; and give the minimum pre-period length you would require.

(3) Estimation: write out the difference-in-differences or event-study specification with geography and week fixed effects, and cluster the standard errors appropriately; if you choose a linear mixed-effects model instead, specify which terms are fixed (such as treatment × post and week effects) and which are random (such as geography intercepts and/or slopes), and justify that choice.

(4) Threats: explain how you would deal with interference or spillovers (for instance, by excluding adjacent buffer geographies), anticipation effects, and seasonality.

(5) Staggered adoption: if other geographies adopt the feature later, name an estimator that is robust to staggered rollout timing—such as Sun–Abraham or Callaway–Sant’Anna—and outline how you would implement it.

(6) Reporting: define the primary metric (for example, orders per active user), compute the ATT with a 95% confidence interval, and interpret a hypothetical +3% lift.

Overview: This question tests skill in causal inference and panel-data experimentation, covering control selection, pre-trend validation, difference-in-differences and event-study estimation, mixed-effects modeling, handling interference and seasonality, staggered-adoption inference, and reporting treatment effects.