← Meta Interview Insights

Meta·Data Scientist·Technical Phone Screen·Intermediate

Intermediate
May 2026

Summary

Stats-heavy technical screen for a Data Scientist role at Meta. The whole thing was built around probability and hypothesis testing in a social media analytics context, which felt pretty focused but also a bit relentless.

Questions Asked (3)

Q1

If the probability that a single user comment is positive is p, what is the probability that two independent comments are both positive?

Product Analytics & Metrics
Author's notes

Straightforward multiplication rule stuff.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the independence assumption and the definition of 'positive' for each comment. Then apply the multiplication rule for independent events to derive the probability as p^2, and briefly discuss the implications and potential pitfalls.

Pro tip: Mention that in real-world data, comments are rarely independent due to user behavior or topic clustering, so the independence assumption is a simplification. This shows you understand the limitations and can think critically about model assumptions.

1. Clarify assumptions

Confirm that the two comments are independent and that each has the same probability p of being positive. Also, define what 'positive' means in this context (e.g., sentiment).

2. Apply probability rule

For independent events, the probability of both occurring is the product of their individual probabilities. So, P(both positive) = p * p = p^2.

3. Consider edge cases

Discuss what happens if p=0 or p=1, and note that the result is always between 0 and 1. Also, mention that if independence does not hold, the calculation would be different.

4. Relate to business context

Explain how this simple probability might be used in practice, such as estimating the likelihood of two positive comments in a row, and discuss potential applications in A/B testing or sentiment analysis.

Key Points to Mention

  • Independence assumption and its implications
  • Multiplication rule for independent events
  • Definition of 'positive' comment
  • Edge cases (p=0, p=1)
  • Real-world limitations (e.g., non-independence due to user behavior)
  • Potential applications in product analytics

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

Q2

Assuming each comment response is independent and identically distributed, if a user's first three responses are positive, what is the probability their fourth response is also positive?

Product Analytics & MetricsA/B Testing & Experimentation
Author's notes

This is where I fumbled a bit.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, clarify that under the i.i.d. assumption, the outcomes are independent, so past responses provide no information about the fourth. Then, state that the probability is simply the underlying probability of a positive response, p, and note that without additional information, p cannot be determined from the given data.

Pro tip: Don't fall for the gambler's fallacy—emphasize that independence means the first three responses are irrelevant. Also, mention that in real-world scenarios, responses are rarely i.i.d., so you'd need to check for user-level correlation or use a hierarchical model.

1. Clarify the assumption

Restate that each response is independent and identically distributed, meaning no dependence between responses.

2. Identify the probability

Recognize that the probability of a positive response is a constant p for each response, including the fourth.

3. Apply independence

Conclude that the first three responses do not affect the fourth, so the conditional probability equals p.

4. Address unknown p

State that p is not given, so the numerical probability cannot be computed without additional information.

5. Discuss real-world implications

Mention that in practice, i.i.d. may not hold, and you would need to model user-level effects or use Bayesian methods.

Key Points to Mention

  • Independence implies no memory: P(X4=1 | X1,X2,X3) = P(X4=1) = p.
  • Identically distributed means the same p applies to all responses.
  • The value of p is unknown; it could be estimated from historical data if available.
  • Gambler's fallacy: assuming a positive streak makes a negative more likely is incorrect under i.i.d.
  • In real data, responses from the same user are often correlated, violating i.i.d.
  • If p is unknown, a Bayesian approach with a prior could be used to estimate it.

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

Q3

Model A returns a positive response 80% of the time and Model B returns one 90% of the time. At the 5% significance level, can you conclude Model B is better? Walk through the statistical test, your null and alternative hypotheses, the test statistic, and the decision rule.

A/B Testing & ExperimentationProduct Analytics & Metrics
Author's notes

This was the meat of the whole interview.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Clarify that the question lacks sample sizes, so you must state assumptions (e.g., equal sample sizes) and use a two-proportion z-test. Walk through hypotheses, test statistic, and decision rule, then conclude based on the computed p-value or critical value.

Pro tip: Always ask about sample size and whether the 80% and 90% are from independent samples; without this, the test is underdetermined. Mention that practical significance may differ from statistical significance.

1. State hypotheses

Define null hypothesis H0: p_A = p_B (or p_B - p_A = 0) and alternative H1: p_B > p_A (one-sided) at α = 0.05.

2. Check assumptions and compute test statistic

Assume independent random samples and large enough sample sizes for normal approximation. Compute pooled proportion and z-statistic: z = (p̂_B - p̂_A) / sqrt(p̂(1-p̂)(1/n_A + 1/n_B)).

3. Determine critical value or p-value

For one-sided α = 0.05, critical z = 1.645. Reject H0 if z > 1.645 or if p-value < 0.05.

4. Make decision and interpret

Compare test statistic to critical value. If significant, conclude Model B is better; otherwise, fail to reject H0. Discuss practical significance and limitations.

Key Points to Mention

  • Two-proportion z-test for independent samples
  • One-sided vs two-sided test (question implies one-sided: 'better')
  • Pooled proportion under null hypothesis
  • Sample size requirement for normal approximation
  • Significance level α = 0.05 and critical value 1.645
  • Practical significance vs statistical significance

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