← Meta Interview Insights

Meta·Data Scientist·Technical Phone Screen·Senior

Senior
May 2026

Summary

Brutal onsite question at Meta for a DS role. One massive open-ended case covering metric design, observational analysis, experiment design, and long-term value quantification, all from two tables. The kind of question where you realize halfway through that you've been answering a slightly different question than they asked.

Questions Asked (1)

Q1

You have two event tables, one for feed views (with viewer, post, relationship type, duration, timestamp) and one for post reactions (like, comment, reshare, follow, hide, report). Your hypothesis is that content from friends drives more social engagement than content from unconnected sources. Design a full analysis end-to-end: define your metrics, validate the hypothesis observationally, design an experiment to test expanding unconnected content, and quantify the value of unconnected exposure even if near-term engagement is lower.

A/B Testing & ExperimentationProduct Analytics & MetricsData Modeling
Author's notes

This question ate me alive a little.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining clear metrics for engagement and exposure, then use observational data to compare engagement rates across relationship types while controlling for confounders. Design an A/B test to measure the causal impact of increasing unconnected content, and finally quantify long-term value through metrics like user retention and network growth.

Pro tip: Acknowledge that near-term engagement may not capture long-term value; propose measuring downstream metrics like user retention and content discovery to show maturity. Also, emphasize the importance of guardrail metrics to ensure user experience isn't harmed.

1. Define Metrics and Hypotheses

Clearly define engagement metrics (e.g., reaction rate per view, average duration) and exposure metrics (e.g., proportion of unconnected content). State the hypothesis that friend content drives higher engagement.

2. Observational Analysis

Analyze historical data to compare engagement rates for friend vs. unconnected content, using stratification or regression to control for confounders like post type, time of day, and user activity.

3. Experiment Design

Design an A/B test where treatment group sees increased unconnected content. Define randomization unit (e.g., user), sample size, duration, and primary/secondary metrics including guardrails.

4. Quantify Long-Term Value

Measure long-term effects through metrics like user retention, network growth, and content diversity. Use holdout groups or long-term tracking to assess value beyond immediate engagement.

Key Points to Mention

  • Confounder control: Use propensity score matching or regression to isolate relationship type effect.
  • Metric definitions: Distinguish between engagement rate (reactions/views) and total engagement; consider duration as quality signal.
  • Experiment pitfalls: Discuss novelty effects, network effects, and SUTVA violations; propose cluster randomization if needed.
  • Long-term value: Propose measuring user lifetime value, retention curves, and ecosystem health metrics.
  • Guardrail metrics: Monitor hide/report rates and user satisfaction to avoid negative experiences.
  • Business impact: Tie findings to strategic goals like increasing time spent or user growth.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.