← Meta Interview Insights

Meta·Data Scientist·Technical Phone Screen·Senior

Senior
May 2026

Summary

Meta data scientist interview with a pretty focused stats question around ad measurement. One question but it had a lot of layers, and the Bayesian twist at the end is where things got interesting.

Questions Asked (1)

Q1

You have conversion counts and exposure numbers for a test group (saw the ad) and a control group (didn't). How do you estimate lift and test whether it's statistically significant? How would you calculate the sample size needed to detect a 5% lift at 90% power? And if a stakeholder asks for a Bayesian framing instead, how do you reframe the whole thing?

A/B Testing & ExperimentationProduct Analytics & MetricsStakeholder Management
Author's notes

Three questions bundled into one, which I didn't fully appreciate until I was halfway through the frequentist part and they asked about sample size.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining lift as the relative difference in conversion rates between test and control, then test significance using a two-proportion z-test or chi-square test. For sample size, use the standard formula for two proportions with 90% power and 5% significance, and for Bayesian framing, shift to posterior distributions and probability of superiority.

Pro tip: Always clarify the baseline conversion rate and whether the 5% lift is relative or absolute, as this drastically changes sample size. Also, mention that Bayesian methods can provide more intuitive results for stakeholders but require prior specification.

1. Define metrics and hypotheses

Clarify conversion rate, lift definition (relative vs absolute), and set up null and alternative hypotheses for frequentist testing.

2. Estimate lift and test significance

Compute conversion rates for test and control, calculate lift, and perform a two-proportion z-test to get a p-value and confidence interval.

3. Calculate sample size

Use the formula for two proportions: n = (Zα/2 + Zβ)^2 * (p1(1-p1) + p2(1-p2)) / (p2-p1)^2, with p2 = p1 * 1.05 for a 5% relative lift.

4. Reframe in Bayesian terms

Model conversion rates with Beta priors, update with data to get posteriors, and compute the probability that test > control or the expected lift distribution.

5. Communicate results to stakeholders

Explain frequentist and Bayesian results in business terms, highlighting uncertainty and decision-making implications.

Key Points to Mention

  • Two-proportion z-test and its assumptions (independence, large sample)
  • Sample size formula components: significance level (α), power (1-β), baseline conversion rate, and minimum detectable effect
  • Bayesian A/B testing: Beta-Binomial model, priors, posteriors, and probability of superiority
  • Difference between relative and absolute lift and its impact on sample size
  • Confidence intervals vs credible intervals for uncertainty quantification
  • Practical significance vs statistical significance and business impact

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