Diagnose 10–11% usage drop across geos

Quick Overview

This question evaluates causal inference and experimentation competencies including metric diagnosis, identification of confounders, experimental design, quasi-experimental reasoning, statistical power considerations, and communication of findings within the Analytics & Experimentation domain for a Data Scientist role.

Diagnose 10–11% usage drop across geos

Company: Google

Role: Data Scientist

Category: Analytics & Experimentation

Difficulty: medium

Interview Round: Technical Screen

US usage is down 10% and Mexico is down 11%. List plausible confounders (seasonality, pricing, outages, marketing mix, competitor moves, feature rollouts, macro). Propose an experiment to isolate causality: define the unit of randomization (user or geo), exposure and segmentation, primary metric and guardrails, duration and power, and pre-registration. If an experiment is infeasible, outline a difference-in-differences or synthetic control with explicit identification assumptions and diagnostics. Specify which covariates to stratify or control for, how you would slice results, and how you would present findings and confidence intervals to stakeholders along with remediation options.

Quick Answer: This question evaluates causal inference and experimentation competencies including metric diagnosis, identification of confounders, experimental design, quasi-experimental reasoning, statistical power considerations, and communication of findings within the Analytics & Experimentation domain for a Data Scientist role.

|Home/Analytics & Experimentation/Google
Google logo
Google
Oct 13, 2025, 9:49 PM
mediumData ScientistTechnical ScreenAnalytics & Experimentation
6
0

US usage is down 10% and Mexico is down 11%. List plausible confounders (seasonality, pricing, outages, marketing mix, competitor moves, feature rollouts, macro). Propose an experiment to isolate causality: define the unit of randomization (user or geo), exposure and segmentation, primary metric and guardrails, duration and power, and pre-registration. If an experiment is infeasible, outline a difference-in-differences or synthetic control with explicit identification assumptions and diagnostics. Specify which covariates to stratify or control for, how you would slice results, and how you would present findings and confidence intervals to stakeholders along with remediation options.

Loading comments...