← PayPal Interview Insights

PayPal·Data Scientist·Technical Phone Screen·Senior

SeniorPrefer not to say
Jul 2026

Summary

PayPal fraud data science interview that was basically a deep dive on p-values and experimentation. More rigorous than I expected for what was framed as a stats refresher.

Questions Asked (4)

Q1

Give a precise definition of a p-value. What does it actually mean, and what are common misconceptions about what it does NOT mean?

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

I started with the textbook definition and then fumbled a bit trying to articulate the 'does not mean' part without sounding like I was reciting a Wikipedia warning label.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start with a precise, technical definition of a p-value, then clearly explain its interpretation in the context of hypothesis testing. Follow by addressing common misconceptions, emphasizing what a p-value does not mean, and connect it to practical implications in A/B testing and product analytics.

Pro tip: Demonstrate maturity by acknowledging the limitations of p-values and mentioning complementary metrics like effect size and confidence intervals, which are crucial for decision-making in industry settings.

1. Precise Definition

State that a p-value is the probability of observing data at least as extreme as the observed data, assuming the null hypothesis is true. Emphasize that it is a conditional probability, not the probability that the null hypothesis is true.

2. Correct Interpretation

Explain that a small p-value indicates that the observed data are unlikely under the null hypothesis, providing evidence against it. Clarify that it does not measure the size or importance of an effect.

3. Common Misconceptions

List what a p-value is not: it is not the probability that the null hypothesis is true, not the probability that results are due to chance, not a measure of effect size or practical significance, and not a definitive proof of any hypothesis.

4. Practical Implications

Discuss how p-values are used in A/B testing to determine statistical significance, but caution against over-reliance. Mention the importance of considering effect size, confidence intervals, and business context for decision-making.

5. Conclusion

Summarize that p-values are a useful tool but must be interpreted carefully and in conjunction with other statistical measures and domain knowledge.

Key Points to Mention

  • Definition: Probability of observing data at least as extreme as the observed, given the null hypothesis is true.
  • Misconception: p-value is not the probability that the null hypothesis is true.
  • Misconception: p-value is not the probability that the results are due to chance.
  • Misconception: p-value does not indicate the size or practical significance of an effect.
  • Context: In A/B testing, p-values help assess statistical significance but should be complemented with effect size and confidence intervals.
  • Limitations: p-values are sensitive to sample size and do not measure the importance or magnitude of an effect.

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

Q2

You run an A/B test on a new friction step like an extra OTP verification and get p=0.03. What can you actually conclude, and what else do you need before making a decision?

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

This is where I think I did okay but undersold the business side.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by interpreting the p-value correctly: it indicates a 3% chance of observing such an extreme result if the null hypothesis (no effect) were true, not a 97% chance the effect is real. Then, discuss the practical significance, potential pitfalls like multiple testing or peeking, and the additional information needed—such as effect size, confidence interval, business impact, and guardrail metrics—before making a decision.

Pro tip: Emphasize that statistical significance does not imply practical significance; always consider the confidence interval and the minimum detectable effect to assess whether the observed effect is meaningful for the business.

1. Interpret the p-value correctly

Explain that p=0.03 means there is a 3% probability of seeing the observed result (or more extreme) if the null hypothesis is true. It does not measure the probability that the null is true or the size of the effect.

2. Assess statistical validity

Check whether the test was designed properly: was the sample size determined in advance? Were there multiple comparisons or peeking? Is the test adequately powered? These factors affect the reliability of the p-value.

3. Evaluate practical significance

Look at the effect size and its confidence interval. Determine if the observed change is large enough to matter for the business, considering the minimum detectable effect and the cost of the friction step.

4. Consider business metrics and guardrails

Examine secondary metrics and guardrail metrics (e.g., conversion rate, customer satisfaction, fraud rates) to ensure the change doesn't harm other important areas. Also consider segment-level effects.

5. Make a decision with holistic view

Combine statistical evidence, practical significance, and business impact to decide whether to implement, iterate, or abandon the change. Consider the cost of further testing versus the potential gain.

Key Points to Mention

  • p-value interpretation: probability of observing data given null hypothesis is true
  • Effect size and confidence interval: magnitude and precision of the estimated effect
  • Multiple testing and peeking: risks of inflated false positive rate
  • Statistical power and sample size: ensuring the test can detect a meaningful effect
  • Practical significance vs statistical significance: business impact of the observed effect
  • Guardrail metrics: ensuring no negative impact on other key metrics

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

Q3

What are at least four common pitfalls when using p-values in practice? Think about things like multiple testing, peeking at results early, p-hacking, and so on.

A/B Testing & ExperimentationRoot Cause Analysis
Author's notes

Got most of these.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by acknowledging that p-values are widely used but often misinterpreted, then list at least four pitfalls with concrete examples. For each pitfall, briefly explain why it's problematic and how to avoid it, tying back to A/B testing and root cause analysis contexts.

Pro tip: Emphasize that p-values only measure evidence against the null hypothesis, not the probability that the null is true or the effect size. Mention that combining p-values with confidence intervals and effect sizes gives a more complete picture.

1. Define p-value and its purpose

Briefly state that a p-value is the probability of observing data at least as extreme as the observed, assuming the null hypothesis is true. Clarify that it is not the probability that the null is true.

2. List common pitfalls

Enumerate at least four pitfalls: multiple testing, peeking/early stopping, p-hacking, and misinterpretation (e.g., equating p-value with effect size or practical significance).

3. Explain each pitfall with examples

For each pitfall, provide a concrete example relevant to A/B testing or root cause analysis, such as running multiple tests without correction or stopping an experiment when p < 0.05.

4. Discuss mitigation strategies

Suggest ways to avoid these pitfalls, such as using Bonferroni or FDR corrections, pre-registering analysis plans, setting a fixed sample size, and using Bayesian methods or effect sizes.

5. Conclude with best practices

Summarize that p-values should be used alongside other metrics like confidence intervals and effect sizes, and always in the context of a well-designed experiment.

Key Points to Mention

  • Multiple testing: inflates false positive rate; correct with Bonferroni, Holm-Bonferroni, or Benjamini-Hochberg procedures.
  • Peeking/early stopping: increases Type I error; use sequential testing or alpha spending functions.
  • P-hacking: manipulating data or analyses to achieve significance; avoid by pre-registering hypotheses and analysis plans.
  • Misinterpretation: p-value is not the probability that the null is true, nor does it measure effect size or practical significance.
  • Optional stopping: another term for peeking, where data is analyzed as it accumulates and stopping when significant.
  • Publication bias: tendency to publish only significant results, leading to distorted literature; relevant in root cause analysis when reviewing past experiments.

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

Q4

How would you adjust your analysis when testing across many regions or segments, when outcomes are rare and delayed like chargebacks that arrive weeks later, or when randomization isn't clean due to spillover effects?

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

Three scenarios in one question is a lot.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer around the three challenges—many regions/segments, rare and delayed outcomes, and spillover—and for each, describe how you would adjust your experimental design, analysis, and decision-making. Emphasize practical trade-offs and the importance of aligning with business goals while maintaining statistical rigor.

Pro tip: Mention that you would pre-register the analysis plan and simulate the impact of delays and spillover to set expectations with stakeholders, showing you think ahead about operational constraints.

1. Clarify the goal and constraints

Start by understanding the business objective, the definition of success, and any operational constraints (e.g., how long you can wait for results). This shapes whether you prioritize speed or accuracy.

2. Adjust design for many regions/segments

Use stratified randomization or block randomization by region/segment to ensure balance. Consider hierarchical models or meta-analysis to pool results while accounting for heterogeneity.

3. Handle rare and delayed outcomes

For rare events, use methods like survival analysis or negative binomial models to handle overdispersion. For delays, consider sequential testing with interim analyses or Bayesian methods that update as data arrives, and account for censoring.

4. Mitigate spillover and interference

Detect spillover via cluster randomization or switchback designs. Use techniques like difference-in-differences or instrumental variables if randomization is compromised. Consider cluster-level analysis or spatial models.

5. Synthesize and decide

Combine evidence across segments using appropriate weighting, and assess practical significance. Communicate uncertainty and recommend actions based on the totality of evidence.

Key Points to Mention

  • Stratified randomization and hierarchical modeling for regional/segment heterogeneity
  • Survival analysis or negative binomial models for rare events
  • Sequential testing or Bayesian updating for delayed outcomes
  • Cluster randomization, switchback designs, or difference-in-differences for spillover
  • Pre-registration and simulation to set expectations
  • Trade-offs between statistical power, speed, and complexity

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