← Upstart Interview Insights

Upstart·Data Scientist·Technical Phone Screen·Senior

SeniorPrefer not to say
May 2026Remote

Summary

Interviewed for a Data Scientist role at Upstart. The question was a meaty causal inference design problem centered on ride-share driver supply, and it pushed pretty hard on experimental design fundamentals. Not a warm-up round.

Questions Asked (1)

Q1

Design an experiment to measure whether surge-pricing push notifications sent to drivers actually increase supply at the airport. What primary and secondary metrics would you track, and how would you handle spillover effects onto drivers who weren't part of the treatment group?

A/B Testing & ExperimentationProduct Analytics & Metrics
Author's notes

This one took me a minute to get my footing.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining the experiment's goal and unit of randomization, then outline primary and secondary metrics with clear definitions. Address spillover by considering cluster randomization or measuring interference, and propose methods to isolate the treatment effect.

Pro tip: Use a switchback or cluster-randomized design to mitigate spillover, and pre-register your analysis plan to avoid p-hacking. Also, consider using a difference-in-differences approach if randomization is compromised.

1. Define Objective and Hypothesis

Clearly state the hypothesis: surge-pricing push notifications increase driver supply at the airport. Specify the target population (e.g., drivers who frequently serve the airport) and the treatment (notification with surge pricing).

2. Choose Randomization Unit and Design

Decide whether to randomize at the driver level or use a cluster/switchback design to handle spillover. Consider geographic or temporal clustering if drivers interact.

3. Select Primary and Secondary Metrics

Primary: number of drivers entering the airport geofence or completed trips from airport per unit time. Secondary: driver acceptance rate, time to accept, earnings per driver, and overall airport wait times.

4. Address Spillover Effects

Use cluster randomization (e.g., by city or time blocks) to minimize contamination. Alternatively, measure spillover by tracking control drivers' behavior and adjust using techniques like difference-in-differences or instrumental variables.

5. Analyze and Validate

Conduct power analysis, run the experiment, and analyze with appropriate statistical tests. Check for novelty effects, seasonality, and ensure results are robust with sensitivity analyses.

Key Points to Mention

  • Randomization unit: driver vs. cluster vs. switchback
  • Primary metric: driver supply at airport (e.g., number of drivers in geofence, trips completed)
  • Secondary metrics: acceptance rate, time to accept, driver earnings, airport wait times
  • Spillover mitigation: cluster randomization, switchback design, or measuring interference
  • Statistical power and sample size calculation
  • Potential confounders: time of day, day of week, weather, events

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