Compare Bayesian and frequentist decisions

Quick Overview

This question evaluates understanding of Bayesian inference versus frequentist hypothesis testing, decision theory under asymmetric loss, sequential monitoring/optional stopping, and prior elicitation applied to A/B testing within the Statistics & Math domain.

Compare Bayesian and frequentist decisions

Company: Meta

Role: Data Scientist

Category: Statistics & Math

Difficulty: medium

Interview Round: Onsite

For a binary KPI with Beta(1,1) priors, you observe 10,000 users per arm and 515 vs. 500 conversions. a) Compute the posterior for each arm and the posterior probability that lift>0. b) Define a decision rule based on expected loss with asymmetric costs. c) Contrast the decision with a frequentist test under optional stopping and discuss operational implications (e.g., communicating posterior vs. p‑value, alpha spending). d) How would you set or elicit priors to avoid over‑optimism?

Quick Answer: This question evaluates understanding of Bayesian inference versus frequentist hypothesis testing, decision theory under asymmetric loss, sequential monitoring/optional stopping, and prior elicitation applied to A/B testing within the Statistics & Math domain.

|Home/Statistics & Math/Meta
Meta logo
Meta
Oct 13, 2025, 9:49 PM
mediumData ScientistOnsiteStatistics & Math
10
0

A/B Test With Beta–Binomial Posteriors and Decision-Making Under Asymmetric Costs

You ran a two-arm A/B test on a binary KPI with independent Beta(1, 1) priors for each arm. Each arm saw 10,000 users. You observed 515 conversions in arm A and 500 in arm B.

Assume "lift > 0" means that the treatment arm A has a higher true conversion rate than control arm B: Δ = p_A − p_B.

Answer the following:

(a) Compute the posterior distribution for each arm and the posterior probability that lift > 0.

(b) Define a decision rule that minimizes expected loss when the costs of a false positive (shipping a worse variant) and a false negative (failing to ship a better variant) are asymmetric.

(c) Contrast the resulting Bayesian decision with a frequentist two-proportion test, especially under optional stopping. Discuss operational implications (e.g., communicating posterior vs. p‑value, sequential monitoring/alpha spending).

(d) How would you set or elicit priors to avoid over‑optimism in future experiments?

Loading comments...