Explain Simpson’s Paradox and Its Causes with Example
Quick Overview
Evaluates Simpson's paradox through definition, confounding, subgroup imbalance, and a numeric reversal example. Strong answers compare stratified and aggregate rates and explain when each comparison matters.
Explain Simpson’s Paradox and Its Causes with Example
Company: Google
Role: Data Scientist
Category: Statistics & Math
Difficulty: medium
Interview Round: Technical Screen
##### Scenario
Interview question on Simpson’s paradox.
##### Question
Define Simpson’s paradox, explain how data imbalance or confounding can create it, and provide a concrete example.
##### Hints
Aggregation vs. stratification effects are key.
Overview: Evaluates Simpson's paradox through definition, confounding, subgroup imbalance, and a numeric reversal example. Strong answers compare stratified and aggregate rates and explain when each comparison matters.
Explain Simpson’s Paradox and Its Causes with Example
Google
Jul 12, 2025
mediumData ScientistTechnical ScreenStatistics & Math
15
0
Simpson's Paradox: Definition, Cause, and Example
Demonstrate your understanding of Simpson's paradox in a statistics or analytics interview.
Define the paradox, explain how imbalance or confounding can create it, and provide a concrete numeric example where a relationship holds within each group but reverses after aggregation.
Constraints & Assumptions
Use stratified and aggregated rates in the example.
Identify the confounding or stratifying variable.
Explain that the paradox is about weighted averages, not arithmetic error.
Discuss how to decide which comparison is relevant.
Clarifying Questions to Ask Guidance
What exposure, outcome, and subgroup variable are being compared?
Is the subgroup variable a confounder, mediator, collider, or segmentation variable?
Is the business decision about subgroup-level or aggregate performance?
Are subgroup sample sizes very different across comparison groups?
Part 1 - Definition
Define Simpson's paradox in your own words.
What This Part Should Cover Guidance
State that a trend appearing within subgroups can reverse or disappear when data are aggregated.
Explain that aggregation changes subgroup weights.
Mention that confounding often drives the reversal.
Part 2 - Why It Happens
Explain how imbalance or confounding creates the paradox.
What This Part Should Cover Guidance
Show that aggregate rates are weighted averages of subgroup rates.
Explain how one group can be overrepresented in harder or easier subgroups.
Connect the confounder to both the exposure and outcome.
Warn against interpreting aggregate results without stratification.
Part 3 - Numeric Example
Provide a concrete numeric example.
What This Part Should Cover Guidance
Use two treatments, products, or cohorts and at least two subgroups.
Show the rate for each group within each subgroup.
Show the aggregate rate and the reversal.
Explain which comparison is more appropriate and why.
Follow-up Questions Guidance
How would you detect Simpson's paradox in an experiment dashboard?
What is the difference between confounding and mere segmentation?
When might the aggregate result still be the right decision metric?