← Meta Interview Insights

Meta·Data Scientist·Technical Phone Screen·Intermediate

Intermediate
Apr 2026

Summary

Meta data scientist interview with a probability/Bayesian question that felt more like a stats exam than a job interview. One question, pretty focused, no fluff.

Questions Asked (1)

Q1

A platform's fake-account detector has a 2% base rate of fake accounts, flags 90% of real fakes, and incorrectly flags 5% of legitimate accounts. If an account gets flagged, what's the actual probability it's fake? Walk through your math.

Product Analytics & MetricsRoot Cause Analysis
Author's notes

This is pure Bayes theorem and I knew that immediately, but I still fumbled the setup for a second because the base rate being so low felt counterintuitive.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Recognize this as a Bayesian probability problem and set up the calculation using the base rate, true positive rate, and false positive rate. Compute the probability of being fake given a flag by dividing the probability of a true positive by the total probability of a flag (true positives + false positives). Clearly explain each step and interpret the result in the context of the platform's detection system.

Pro tip: After computing the exact probability, mention that this illustrates the base rate fallacy and discuss the practical implications for the platform, such as the need for additional verification steps or adjusting the threshold to balance precision and recall.

1. Identify given probabilities

Extract the base rate (2% fake), true positive rate (90% of fakes flagged), and false positive rate (5% of legitimate flagged).

2. Define the events and goal

Let F be the event that an account is fake, and A be the event that it is flagged. We need P(F|A).

3. Apply Bayes' theorem

Use the formula P(F|A) = P(A|F) * P(F) / [P(A|F)*P(F) + P(A|F^c)*P(F^c)].

4. Plug in numbers and compute

Calculate numerator: 0.9 * 0.02 = 0.018. Denominator: 0.018 + 0.05 * 0.98 = 0.018 + 0.049 = 0.067. Then P(F|A) = 0.018 / 0.067 ≈ 0.2687 or 26.87%.

5. Interpret the result

Explain that despite the high true positive rate, the low base rate means only about 27% of flagged accounts are actually fake, highlighting the base rate fallacy.

Key Points to Mention

  • Bayes' theorem and conditional probability
  • Base rate fallacy and its impact on interpreting test results
  • True positive rate (sensitivity) vs. false positive rate
  • Precision (positive predictive value) and its importance in product metrics
  • Practical implications for the platform, such as threshold tuning or additional checks
  • Clear communication of the result to non-technical stakeholders

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