← Walmart Labs Interview Insights

Walmart Labs·Data Scientist·Technical Phone Screen·Intermediate

IntermediatePrefer not to say
Jun 2026Remote

Summary

Interviewed for a data scientist role on the growth analytics team at Walmart Labs. The session was heavily focused on causal inference methodology, specifically PSM, which I hadn't drilled as hard as I probably should have.

Questions Asked (3)

Q1

What is Propensity Score Matching and when would you actually use it in a business context?

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

I knew the textbook definition but stumbled when they pushed for concrete business scenarios.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining Propensity Score Matching (PSM) as a causal inference technique that balances treatment and control groups on observed covariates by matching units with similar propensity scores. Then explain when to use it in business: when randomized experiments are infeasible or unethical, and you need to estimate causal effects from observational data. Finally, illustrate with a concrete Walmart Labs example, such as evaluating a loyalty program's impact on customer spend, and discuss limitations like unobserved confounding.

Pro tip: Emphasize that PSM only controls for observed confounders, so it's not a substitute for randomization; always check covariate balance after matching and consider sensitivity analysis for unobserved confounding. Mention that in tech companies, PSM is often used for quasi-experiments when A/B tests are impossible, but be ready to discuss alternatives like difference-in-differences or synthetic control.

1. Define PSM clearly

Explain that PSM estimates the probability of receiving treatment given observed covariates (the propensity score) and then matches treated and control units with similar scores to create comparable groups.

2. Explain the goal: causal inference from observational data

Highlight that PSM aims to reduce selection bias by balancing covariates, allowing you to estimate the average treatment effect on the treated (ATT) when randomization isn't possible.

3. Describe when to use it in business

Give scenarios where experiments are impractical: e.g., evaluating a marketing campaign that was rolled out to a specific segment, assessing the impact of a loyalty program, or measuring the effect of a feature adoption when users self-select.

4. Provide a concrete Walmart Labs example

Illustrate with a relevant case: e.g., estimating the effect of a personalized recommendation feature on purchase frequency, where some customers were exposed due to prior behavior, and you match on demographics, past purchase history, and engagement metrics.

5. Discuss limitations and best practices

Mention that PSM only addresses observed confounding, requires sufficient overlap, and needs balance checks. Suggest combining with sensitivity analysis or other methods like difference-in-differences for robustness.

Key Points to Mention

  • Propensity score definition: probability of treatment assignment given covariates.
  • Matching methods: nearest neighbor, caliper, stratification, or weighting (IPW).
  • Causal estimand: typically ATT (average treatment effect on the treated).
  • Balance diagnostics: standardized mean differences, variance ratios, and love plots.
  • Common support/overlap assumption: treated and control units must have similar propensity scores.
  • Limitations: unobserved confounding, model dependence, and sensitivity to matching algorithm choices.

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

Q2

What assumptions need to hold for PSM to give you unbiased causal estimates?

A/B Testing & ExperimentationTechnical Trade-offs
Author's notes

Got ignorability out fast, then common support.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying that PSM (Propensity Score Matching) aims to estimate causal effects by balancing observed covariates between treated and control groups. Then, systematically outline the key assumptions: conditional independence (unconfoundedness), overlap (common support), and correct specification of the propensity score model. Finally, discuss how violations of these assumptions can bias estimates and mention sensitivity checks.

Pro tip: Emphasize that PSM only controls for observed confounders; unobserved confounding can still bias results, so it's crucial to discuss sensitivity analysis or alternative methods like instrumental variables when unobserved confounding is a concern.

1. Define PSM and its goal

Briefly explain that PSM matches treated and control units with similar propensity scores to mimic randomization, aiming to estimate the average treatment effect on the treated (ATT).

2. State the core assumptions

List the key assumptions: conditional independence (no unmeasured confounding), overlap (common support), and stable unit treatment value assumption (SUTVA).

3. Explain each assumption in detail

For each assumption, describe what it means and why it's necessary for unbiased causal estimates. For example, conditional independence requires that treatment assignment is independent of potential outcomes given observed covariates.

4. Discuss consequences of violations

Explain how violating these assumptions leads to bias: unmeasured confounding biases estimates, lack of overlap leads to extrapolation, and SUTVA violations cause interference.

5. Mention practical checks and limitations

Suggest ways to assess assumptions (e.g., balance checks, sensitivity analysis) and acknowledge that PSM cannot address unobserved confounding.

Key Points to Mention

  • Conditional independence (unconfoundedness): no unmeasured confounders.
  • Overlap (common support): both treated and control units have a positive probability of receiving treatment.
  • SUTVA: no interference between units and consistent treatment definition.
  • Correct specification of the propensity score model (e.g., logistic regression with relevant covariates).
  • PSM estimates ATT, not ATE, unless additional assumptions hold.
  • Sensitivity analysis for unobserved confounding (e.g., Rosenbaum bounds).

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

Q3

After running PSM, how do you check that the matching actually worked? Walk me through your balance diagnostics.

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

Mentioned standardized mean differences and love plots.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by explaining that after PSM, you assess balance by comparing covariate distributions between treated and control groups before and after matching, using standardized mean differences (SMD) as the primary metric. Then describe complementary diagnostics like variance ratios, propensity score overlap, and sensitivity analysis to ensure robustness. Emphasize that balance checks are iterative—if imbalances persist, you refine the model or matching algorithm.

Pro tip: Always report both pre- and post-matching SMDs in a Love plot; this visually demonstrates improvement and helps stakeholders quickly grasp matching quality. Also, don't just rely on SMD—check variance ratios and consider the effective sample size after matching to avoid overfitting.

1. Compute Standardized Mean Differences (SMD)

Calculate SMD for each covariate before and after matching. A common rule of thumb is that SMD < 0.1 indicates adequate balance.

2. Assess Variance Ratios

Compare the variance of each covariate between treated and control groups post-matching. Ratios close to 1 (e.g., between 0.5 and 2) suggest balanced distributions.

3. Visualize Balance with Love Plots

Create a Love plot to display SMDs for all covariates before and after matching. This helps identify which covariates remain imbalanced.

4. Check Propensity Score Overlap and Common Support

Examine the distribution of propensity scores in both groups to ensure sufficient overlap. Discard observations outside common support if necessary.

5. Conduct Sensitivity Analysis

Perform sensitivity analysis (e.g., Rosenbaum bounds) to assess how robust results are to unmeasured confounding. This adds credibility to causal claims.

Key Points to Mention

  • Standardized mean differences (SMD) and the 0.1 threshold
  • Variance ratios and their interpretation
  • Love plots for visual balance assessment
  • Propensity score overlap and common support
  • Effective sample size after matching
  • Sensitivity analysis for unmeasured confounding

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