Evaluate a One-Day Delivery Launch with Difference-in-Differences
Company: Amazon
Role: Data Scientist
Category: Analytics & Experimentation
Difficulty: medium
Interview Round: Technical Screen
# Evaluate a One-Day Delivery Launch with Difference-in-Differences
A business is launching one-day delivery and wants to estimate its causal effect. Propose an evaluation design and explain when difference-in-differences would be appropriate. Focus on how you would assess parallel trends and what you would do if that assumption is doubtful. The rollout unit, outcome, and treatment timing have not been specified; state what you would clarify and make any working assumptions explicit.
### What a Strong Answer Covers
- A defined treatment, outcome, unit, and causal estimand for the delivery launch.
- A suitable comparison group and the identifying assumptions behind difference-in-differences.
- Pre-period trend and event-study diagnostics without treating nonsignificance as proof.
- A response to anticipation, spillovers, differential shocks, or staggered rollout.
```hint Separate levels from trends
A comparison group can start at a different outcome level and still have a useful untreated trend.
```
### Follow-up Questions
- Why does a statistically insignificant pre-trend test fail to establish parallel trends?
- What changes if different markets receive one-day delivery at different times?
Overview: Practice causal evaluation of one-day delivery using difference-in-differences, parallel-trend diagnostics, and staggered-rollout considerations.
Evaluate a One-Day Delivery Launch with Difference-in-Differences
A business is launching one-day delivery and wants to estimate its causal effect. Propose an evaluation design and explain when difference-in-differences would be appropriate. Focus on how you would assess parallel trends and what you would do if that assumption is doubtful. The rollout unit, outcome, and treatment timing have not been specified; state what you would clarify and make any working assumptions explicit.
What a Strong Answer Covers Guidance
A defined treatment, outcome, unit, and causal estimand for the delivery launch.
A suitable comparison group and the identifying assumptions behind difference-in-differences.
Pre-period trend and event-study diagnostics without treating nonsignificance as proof.
A response to anticipation, spillovers, differential shocks, or staggered rollout.
Follow-up Questions Guidance
Why does a statistically insignificant pre-trend test fail to establish parallel trends?
What changes if different markets receive one-day delivery at different times?