← Amazon Interview Insights

Amazon·Data Scientist·Technical Phone Screen·Intermediate

Intermediate
Jul 2026Remote

Summary

Phone screen for a Data Scientist role at Amazon, focused pretty heavily on causal inference. One meaty technical question that took up most of the time.

Questions Asked (1)

Q1

Explain Propensity Score Matching: what it is, what assumptions it relies on, how you'd implement it step by step, and how you'd check whether covariate balance was achieved after matching.

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

This question is deceptively wide.

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 for observational data, then clearly state its core assumptions (conditional ignorability, overlap, SUTVA). Walk through a step-by-step implementation: estimate propensity scores, match treated and control units, assess covariate balance, and estimate treatment effects. Emphasize that balance checking is critical and should be done before interpreting results.

Pro tip: At Amazon, where experimentation is central, highlight that PSM is a fallback when randomization isn't possible, but always acknowledge its limitations and suggest sensitivity analysis to test robustness against unmeasured confounding.

1. Define PSM and its purpose

Explain that PSM is a statistical matching technique used to estimate causal effects in observational studies by balancing observed covariates between treated and control groups. It mimics randomization by creating a comparable control group based on propensity scores.

2. State key assumptions

List and briefly explain the assumptions: conditional ignorability (no unmeasured confounding), overlap (common support), and stable unit treatment value assumption (SUTVA). Mention that violation of these can bias results.

3. Outline implementation steps

Describe the process: (1) select covariates, (2) estimate propensity scores (e.g., logistic regression), (3) choose a matching algorithm (nearest neighbor, caliper, etc.), (4) perform matching, (5) check balance, (6) estimate treatment effect on matched sample.

4. Check covariate balance

Explain methods to assess balance: standardized mean differences (SMD) < 0.1, variance ratios, visual inspection (e.g., love plots), and statistical tests (e.g., t-tests). Emphasize that balance should be checked on all covariates used in the model.

5. Estimate treatment effect and sensitivity

After achieving balance, estimate the average treatment effect on the treated (ATT) using matched samples. Conduct sensitivity analysis to assess robustness to unmeasured confounding.

Key Points to Mention

  • Propensity score definition: probability of treatment assignment given observed covariates.
  • Common matching methods: nearest neighbor, caliper, radius, stratification, inverse probability weighting.
  • Balance diagnostics: standardized mean differences, variance ratios, and visual plots.
  • Importance of common support and trimming to avoid extrapolation.
  • Difference between PSM and randomization: PSM only balances observed covariates.
  • Sensitivity analysis (e.g., Rosenbaum bounds) to address unmeasured confounding.

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