##### Scenario Assess the causal impact of providing free shuttle buses on employee participation rates across 1,000+ company sites. ##### Question At what data grain (site-level vs. individual-level) would you run the analysis and why? Start with an OLS specification: write the regression equation, list key controls, and explain how to interpret the shuttle-service coefficient. What limitations does basic OLS have in this context and how would a Difference-in-Differences (DiD) design address them? Design a placebo test to check the DiD identifying assumption. If DiD assumptions fail, outline how you would apply Propensity Score Matching (PSM). Which variables would you include and why? Give a high-level intuition for why Lasso regression performs effective feature selection when estimating propensity scores. Discuss trade-offs of matching with vs. without replacement and how you would run balance checks. After matching, what is the next analytical step (e.g., PSM + DiD) and how would you report results to stakeholders? ##### Hints Cover assumptions, diagnostics, coefficient interpretation, and stakeholder communication; reference causal inference best practices.

Evaluates causal analysis of free shuttle bus impact on employee participation across many sites. Strong answers choose site-time or individual-time grain, write OLS and DiD specifications, control confounding, cluster standard errors, and validate parallel trends with placebo tests.

How do I approach Analytics & Experimentation interview questions?

Analytics & Experimentation questions require understanding of core concepts and practice. PracHub provides solutions with explanations to help you master analytics & experimentation interviews.

What difficulty level is this interview question?

This is a hard difficulty Analytics & Experimentation question, commonly asked during Technical Screen rounds at Meta.

What role is this question designed for?

This question is commonly asked for Data Scientist candidates at Meta during technical interviews.

Analyze Free Shuttle Impact on Employee Participation Rates

You have panel data for more than 1,000 company sites over time. Some sites adopt free shuttle buses at different dates, while others never adopt. The goal is to estimate the causal effect of offering shuttle service on employee participation rates.

Constraints & Assumptions

Treatment is assigned at the site level.
Participation can be measured by site and time, and possibly by individual and time.
Adoption timing varies across sites.
A credible causal design must address site differences and time trends.

Clarifying Questions to Ask Guidance

What is the participation outcome: attendance, commuting mode, event participation, or program enrollment?
What is the time grain: daily, weekly, or monthly?
Are shuttle adoption dates exogenous, or did sites adopt because participation was already changing?
Are individual employee histories available?

Part 1 - Data Grain

Would you analyze at the site level or individual level, and why?

What This Part Should Cover Guidance

Site-time panel as primary because treatment varies by site.
Individual panel as a complementary robustness check if available.
Clustering standard errors at the treatment assignment level.

Part 2 - OLS Baseline

Write an OLS regression equation to estimate the shuttle effect, list key controls, and explain the coefficient.

What This Part Should Cover Guidance

Outcome, treatment indicator, site and time controls, covariates, and interpretation.
Controls for site size, location, job mix, seasonality, local transit, weather, and time trends.

Part 3 - DiD Design

What limitations does basic OLS have, and how would Difference-in-Differences address them?

What This Part Should Cover Guidance

Confounding from site selection and time trends.
Two-way fixed effects or event-study specification.
Parallel trends assumption and staggered adoption concerns.

Part 4 - Placebo Test

Design a placebo test to assess the DiD identifying assumption.

What This Part Should Cover Guidance

Fake treatment dates, pre-period leads, never-treated controls, or outcomes that should not be affected.
Interpretation of placebo failures.

What a Strong Answer Covers Guidance

A strong answer matches the data grain to treatment assignment, uses fixed effects and clustered uncertainty, and validates DiD assumptions with pre-trends and placebo tests.

Follow-up Questions Guidance

What if high-growth sites adopted shuttles first?
How would you handle staggered adoption bias?
What if individual employees transfer between sites?