Katkat
Back

Explain p-values and interpret regressions

PayPalData Scientist · Onsite
EasyData ScientistOnsiteStatistics & Math

Last updated

A PayPal Data Scientist onsite statistics rapid-fire covering p-values (plain-language and formal definitions, common misinterpretations, and an A/B-test decision at p=0.03), interpreting linear-regression coefficients under confounding (associational vs causal, 'controlling for' failure modes, and bias-reduction methods), and L1 vs L2 regularization with how to choose the penalty strength. It tests both statistical depth and the ability to communicate to product and technical audiences.

Question

This is a statistics rapid-fire onsite for a Data Scientist role. Answer each part clearly and precisely — first as if explaining to a Product Manager, then for a technical audience.

Part A — p-values

  1. Explain what a p-value is in plain language (PM-friendly).
  2. Give the formal definition of a p-value.
  3. How should a p-value be interpreted, and what are the common misinterpretations to avoid?
  4. If you run an A/B test and obtain p = 0.03 for the primary metric, what decision would you make? What additional context would you request before shipping?

Part B — Linear regression with confounding 5. You fit a linear regression with multiple features and suspect confounding factors exist. How do you interpret each parameter (coefficient)? 6. What does “controlling for other variables” actually mean, and when can that interpretation fail? 7. What is the difference between an associational and a causal interpretation of a coefficient? 8. What checks or approaches would you use to reduce confounding bias?

Part C — L1 vs L2 regularization 9. What are L1 (Lasso) and L2 (Ridge) regularization, and how do they differ in effect? 10. When and why would you prefer one over the other (feature selection, multicollinearity, prediction vs interpretability)? 11. How would you select the regularization strength (λ), and what are the practical tradeoffs?