Measure causal impact of YouTube ads
Company: Google
Role: Data Scientist
Category: Analytics & Experimentation
Difficulty: Medium
Interview Round: Onsite
Estimate the incremental effect of a 6‑week YouTube campaign on weekly online sales.
- Explain why naive OLS of sales on ad spend is biased; list at least three confounders (e.g., seasonality, promotions, targeting) and the likely direction of bias.
- Propose a primary design (geo‑level randomized controlled trial or matched‑market test) and a backup quasi‑experiment (difference‑in‑differences or synthetic control). State identification assumptions and diagnostics.
- Incorporate ad‑stock/lagged effects (e.g., Koyck/geometric decay) and define the estimand: incremental ROAS over and post campaign.
- Compute required sample size for a geo‑experiment given baseline weekly sales μ=100,000, coefficient of variation=0.25, minimal detectable effect=3%, α=0.05, power=0.8; show formulas and approximate number of geos per arm.
- Detail pre‑trend checks, spillover/interference detection, and how you will aggregate heterogeneous treatment effects (by geo size, baseline sales, or audience overlap).
Quick Answer: This question evaluates causal inference and experimental design skills for marketing measurement, including competency in identifying confounders, designing geo-level randomized or quasi-experimental comparisons, modeling ad-stock/lagged effects, conducting power/sample-size calculations, detecting spillovers, and aggregating heterogeneous treatment effects. It is commonly asked to test practical ability to justify identification assumptions and diagnostics in Analytics & Experimentation, assessing both conceptual understanding of causal frameworks and hands-on practical application of experimental and time-series methods.