← Uber Interview Insights

Uber·Data Scientist·Technical Phone Screen·Senior

Senior
Apr 2026

Summary

Uber DS interview that went deep into experimentation design for a two-sided marketplace. The whole session was basically one long question about switchback experiments, and it got technical fast.

Questions Asked (1)

Q1

Design an experiment to measure the causal effect of showing different ETAs on whether a rider session ends in a completed ride request. Walk through why a standard individual-level A/B test breaks down here, and propose an alternative experimental design.

A/B Testing & ExperimentationProduct Analytics & MetricsTechnical Trade-offs
Author's notes

This is the kind of question where you think you know where it's going and then it just keeps going.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining the causal question and the outcome metric, then explain why individual-level randomization fails due to interference between riders and drivers in the marketplace. Propose a cluster-based or switchback design that accounts for interference, and outline how you would analyze the results to estimate the causal effect.

Pro tip: Emphasize that in marketplace experiments, the stable unit treatment value assumption (SUTVA) is violated, and show you understand the trade-off between cluster size and statistical power. Mention that you would validate the design with a pre-experiment power analysis and consider using a difference-in-differences approach if clusters are not perfectly randomized.

1. Clarify the causal question and metric

Define the treatment (different ETAs), the outcome (completed ride request), and the population (riders). State the causal estimand, e.g., the average treatment effect on the treated (ATT) or the average treatment effect (ATE).

2. Explain why individual-level A/B test fails

Discuss interference: a rider's ETA depends on driver availability, which is affected by other riders' requests. Randomizing at the rider level can lead to contamination because treated and control riders compete for the same drivers, violating SUTVA.

3. Propose an alternative design

Suggest a cluster-randomized design (e.g., by city or time window) or a switchback design where treatment is alternated over time within the same market. Explain how this reduces interference and allows estimation of causal effects.

4. Address analysis and power

Describe how to analyze the experiment: use cluster-level summaries, mixed-effects models, or difference-in-differences. Discuss power calculations considering intra-cluster correlation and the number of clusters.

5. Discuss trade-offs and validity

Acknowledge limitations: cluster randomization may reduce power, switchback designs may have carryover effects, and external validity may be limited. Suggest sensitivity checks or complementary analyses.

Key Points to Mention

  • SUTVA violation and interference in marketplace experiments
  • Cluster randomization or switchback design as solutions
  • Intra-cluster correlation and its impact on power
  • Difference-in-differences or fixed effects for analysis
  • Carryover effects and washout periods in switchback designs
  • Trade-off between bias reduction and statistical power

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