Establish causality: commute playlist and driving speed
Company: Google
Role: Data Scientist
Category: Analytics & Experimentation
Difficulty: medium
Interview Round: Technical Screen
A lawyer worries that listening to a "Commute" playlist in a mobile app makes users drive faster. As the DS: a) Define the population, unit of analysis (e.g., user‑trip), treatment/exposure (playlist shown vs. listened), and outcome (average mph) with precise time windows. b) List ≥5 major confounders and how you would measure them. c) Propose a safe RCT (eligibility, gating, safety guardrails/kill switches, metrics, stopping rules). d) If an RCT is infeasible, detail quasi‑experimental strategies (within‑driver fixed effects, staggered adoption DiD, instrumenting with exogenous surfacing changes, RD on a ranking score threshold), including identifying assumptions and falsification tests. e) Handle time‑varying treatment (playlist starts mid‑trip), partial compliance, and non‑drivers; specify data granularity to avoid leakage.
Overview: This question evaluates a candidate's competency in causal inference, experimental design, measurement, and operational safety for product analytics, including precise definition of population and unit of analysis, treatment/exposure and outcome windows, confounder identification, and handling of time‑varying treatment and compliance.
Read the full Google Data Scientist interview experience this question came from