← Attentive Interview Insights

Attentive·Data Scientist·Technical Phone Screen·Intermediate

Intermediate
Jul 2026

Summary

A stats-heavy question from what seemed like a technical screen for a DS role at Attentive. The whole thing centered on one scenario about multiple testing across 30 brands, which sounds manageable until you're actually in it.

Questions Asked (1)

Q1

A company runs an A/B test on a new message-sending method across 30 independent brands, each with a 50/50 control/test split at α = 0.05. Two brands show a significant lift, one shows a significant drop, and the other 27 show nothing. What conclusions can you draw, and how do you handle the multiple testing problem here?

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

My first instinct was to say 'two out of thirty is actually pretty good' and I almost stopped there, which would've been embarrassing.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, acknowledge that with 30 independent tests at α=0.05, we expect about 1.5 false positives by chance alone, so the observed results (2 significant lifts, 1 significant drop) are roughly consistent with noise. Then, discuss methods to control for multiple testing, such as Bonferroni correction or false discovery rate (FDR), and emphasize the need to consider practical significance and business context before drawing conclusions.

Pro tip: Don't just focus on statistical significance; highlight the importance of effect sizes and confidence intervals to assess practical relevance, and suggest pre-registering hypotheses or using hierarchical models to share information across brands.

1. Quantify the multiple testing problem

Calculate the expected number of false positives under the null hypothesis (30 * 0.05 = 1.5) and note that observing 3 significant results is not surprising.

2. Apply multiple testing corrections

Discuss correction methods like Bonferroni (α = 0.05/30 ≈ 0.0017) or Benjamini-Hochberg FDR, and note that under these corrections, none of the results would likely remain significant.

3. Consider dependence and hierarchical structure

Recognize that brands may not be independent; use hierarchical models or meta-analysis to borrow strength across brands and estimate overall effect.

4. Evaluate practical significance and business impact

Even if statistically significant, assess whether the lift/drop is meaningful for the business, and consider the cost of false positives vs. false negatives.

5. Recommend next steps

Suggest replicating the experiment, increasing sample size, or running a follow-up test on the brands that showed signals to confirm findings.

Key Points to Mention

  • Family-wise error rate (FWER) and false discovery rate (FDR) control
  • Bonferroni correction and its conservativeness
  • Benjamini-Hochberg procedure for FDR
  • Expected false positives under the null (1.5 out of 30)
  • Practical significance vs. statistical significance
  • Hierarchical modeling or meta-analysis to account for brand-level variation

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