Characterize and compare transfer-count distributions over time

Quick Overview

This question evaluates a candidate's ability to model zero-inflated, heavy-tailed count data, interpret distributional summaries and percentiles, reason about temporal shifts from retention and fraud dynamics, and identify robust summary statistics and diagnostics for a Data Scientist role.

Characterize and compare transfer-count distributions over time

Company: Meta

Role: Data Scientist

Category: Statistics & Math

Difficulty: medium

Interview Round: Onsite

For a new cohort, let X be each user’s number of P2P transfers in the first 30 days post‑signup. 1) Argue for a plausible distributional family for X (e.g., zero‑inflated negative binomial or lognormal mixture) and justify expected zero‑mass and heavy‑tail behavior. 2) Sketch the likely ordering and approximate locations of mode, median, mean, and 95th percentile for X, explaining why they differ. 3) Describe how this distribution should evolve by day 60 (selection on retention, habit formation, fraud suppression, seasonality), and predict the directional shifts of those four summaries. 4) Recommend two robust executive‑level summaries resilient to heavy tails (e.g., trimmed mean, median‑of‑ratios) and one diagnostic (e.g., QQ plot or tail index) to validate assumptions.

Quick Answer: This question evaluates a candidate's ability to model zero-inflated, heavy-tailed count data, interpret distributional summaries and percentiles, reason about temporal shifts from retention and fraud dynamics, and identify robust summary statistics and diagnostics for a Data Scientist role.

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

P2P Transfer Counts in First 30 Days: Distribution, Summaries, and Evolution

Context: For a new user cohort, define X as each user’s number of peer-to-peer (P2P) transfers completed in the first 30 days post-signup. X is a nonnegative integer count that often has many zeros (inactive users) and a heavy right tail (high-intensity users and potential abuse).

Tasks

  1. Propose a plausible distributional family for X (e.g., zero-inflated negative binomial, hurdle model, or a lognormal mixture). Justify both the expected mass at zero and heavy-tail behavior.
  2. Sketch the likely ordering and approximate locations of the mode, median, mean, and 95th percentile for X, and explain why they differ.
  3. Describe how this distribution should evolve by day 60 considering selection on retention, habit formation, fraud suppression, and seasonality. Predict directional shifts of the mode, median, mean, and 95th percentile.
  4. Recommend two robust, executive-level summaries resilient to heavy tails (e.g., trimmed mean, median-of-ratios) and one diagnostic (e.g., QQ plot or tail index) to validate assumptions.
Loading comments...