Compare Instagram vs. Facebook using causal experiments
Overview
This question evaluates causal inference, experimentation design, metric engineering, and ecosystem-level impact assessment skills for a Data Scientist in the Analytics & Experimentation domain, with emphasis on defining a single-objective OEC that prevents pure cannibalization and on measuring cross-app effects via randomized rollouts.
Compare Instagram and Facebook for consumer time and engagement: a) Define a single-objective OEC that captures healthy cross-app ecosystem value without rewarding pure cannibalization (e.g., weighted time, sessions, and creator interactions with guardrails on churn and ad quality). b) Design a causal measurement plan for a new Instagram feature (e.g., Reels remix) that might shift time from Facebook: choose between user-level vs. geo-level cluster randomization; propose a staggered geo rollout with Facebook holdouts to measure cross-app impact; outline diff-in-diff with pre-trend checks, calendar effects, and cluster-robust SEs; and specify interference mitigation (geo buffers, cross-over suppression). c) Compute back-of-envelope sample size and test duration given baseline Facebook time = 20 min/day, ICC=0.02 at geo level, MDE=1% on ecosystem OEC, power=0.8, alpha=0.05. d) List guardrail metrics (crashes, feed quality, creator DAU, ads LTV) and stopping rules. e) Explain how you would interpret results if Instagram improves OEC but Facebook time drops by 3%, and propose follow-up experiments or pricing changes to preserve ecosystem health.