Decide if subgroup increases imply overall increase

Quick Overview

This question evaluates understanding of aggregation effects (Simpson's paradox), weighted averages, subgroup composition, and the ability to reason about statistical inference and diagnostic checks across strata.

Decide if subgroup increases imply overall increase

Company: TikTok

Role: Data Scientist

Category: Statistics & Math

Difficulty: medium

Interview Round: Technical Screen

If the average daily time spent on TikTok increases within each gender subgroup (male and female), must the overall average increase as well? Provide: (1) a formal statement and proof or counterexample (Simpson’s paradox) with concrete numbers; (2) necessary and sufficient conditions in terms of subgroup means and population weights under which the overall average is guaranteed to increase; (3) how composition shifts (e.g., male share changing) affect the conclusion; (4) a diagnostic you would run on real data to avoid incorrect inferences.

Quick Answer: This question evaluates understanding of aggregation effects (Simpson's paradox), weighted averages, subgroup composition, and the ability to reason about statistical inference and diagnostic checks across strata.

|Home/Statistics & Math/TikTok
TikTok logo
TikTok
Oct 13, 2025, 9:49 PM
mediumData ScientistTechnical ScreenStatistics & Math
5
0

TikTok Time: Subgroup Increases vs Overall Average (Simpson's Paradox)

You are analyzing average daily time spent on TikTok by gender (male, female) across two periods (baseline t0 and follow-up t1). For each gender, the within-group average time increases from t0 to t1.

Answer the following:

  1. Formal statement and proof or counterexample
  • State whether the overall average must increase when each subgroup's average increases.
  • Provide either a proof or a concrete counterexample (Simpson's paradox) with numbers.
  1. Necessary and sufficient conditions
  • Express conditions, in terms of subgroup means and population weights, under which the overall average is guaranteed to increase.
  1. Impact of composition shifts
  • Explain how changes in group shares (e.g., male share changing) affect the conclusion.
  1. Diagnostic on real data
  • Describe a diagnostic you would run on real data to avoid incorrect inferences caused by composition changes.
Loading comments...