Design a Top Dasher experiment with interference

## Experimentation: Evaluate a “Top Dasher” incentive program A delivery platform wants to launch or change a **Top Dasher** program (e.g., priority access to orders, higher ranking, or incentives) intended to improve reliability and coverage. However, this is a marketplace with strong **interference/spillovers**: if some couriers get priority, other couriers and customers may be affected. ### Task 1. Define the **goal** of the program and propose a **metric tree**: - Primary outcome metric(s) - Diagnostic metrics - Guardrails 2. Choose an **experimental design** and justify the **randomization unit** (examples: courier, courier-day, zone, zone-time). 3. Discuss when to use: - Individual randomization - Cluster randomization - **Switchback** experiments - Clustered analysis / variance estimation 4. If using a **switchback** design, list the main threats (carryover, noncompliance, seasonality, novelty) and how you would mitigate them. 5. Explain how you would reason about **false positives vs false negatives** (Type I vs Type II error) for this decision, including which risk is worse and why. ### Output Provide a concrete recommended design (including analysis approach) and how you would communicate trade-offs to stakeholders.