Foundations
- Normal distribution why-everywhere? — CLT: means of samples converge to normal.
- Mean vs median? — median for skewed (salaries!); give example.
- Variance vs std-dev? — spread in squared vs original units.
- Correlation vs causation? — ice-cream vs drowning classic + confounders.
Hypothesis Testing & A/B Tests
- p-value in one line? — probability of data this extreme if null were true; not "H0 probability".
- Type I vs II errors? — false alarm vs miss; power = 1 − β.
- Design an A/B test? — metric, randomisation, duration (power), SRM check, decision rule.
- Peeking problem? — inflates false positives; fix with sequential methods.
Applied Favourites
- Bayes theorem use? — spam filters, medical tests; base-rate example.
- What is power analysis? — sample size for detectable effect.
- Simpson's paradox? — aggregate reversals; segment before concluding.
10 More Rapid-Fire Questions (With Answers)
- Law of large numbers? — Sample mean converges to population mean as n grows — why big samples stabilise.
- CLT in one example? — Average of 30 dice-roll batches looks normal even though rolls are uniform.
- Covariance vs correlation? — Covariance is scaled; correlation (-1..1) is comparable.
- Skewed salary data — which average? — Median; report both + percentiles.
- Outlier detection? — IQR fences or z-score >3; always ask domain before deleting.
- Sampling methods? — Random, stratified (keeps ratios), cluster — match method to population.
- Reading a CI? — "95% CI [4,6]" = procedure captures truth 95% of the time; not per-interval probability.
- Selection bias example? — App-only survey missing offline users — state who is excluded.
- Multiple comparisons fix? — Bonferroni/FDR — more tests, more false hits; adjust thresholds.
- Spurious correlation? — Pool drownings vs ice-cream (summer confounder) — control before claiming.
7-Day Revision Plan
- Day 1–2: definitions aloud (record yourself, 2-min each).
- Day 3–4: 1 project end-to-end narration.
- Day 5: full mock with a friend/mentor.
- Day 6: fix weak answers only.
- Day 7: rest + revise flashcards. Sleep beats cramming.
Annotated Mock Round (Product Analyst, 15 min)
- "Checkout conversion dropped 12% — debug?" → "Segment (device/channel), check SRM + seasonality, then experiment inventory." Structured thinking.
- "p = 0.06 — ship?" → "Not at α=0.05; check power, effect size, cost of wrong ship — statistics serves decisions." Judgment.
- "Explain Bayes to marketing." → "Start with base rate: rare events need strong evidence — mammogram numbers." Teaching = seniority.
- "Peeking at tests daily?" → "Inflates false positives; pre-commit N + use sequential bounds." Rigour.
Frequently Asked Questions
- Is statistics asked in analyst interviews?
Yes — A/B testing + distributions appear in startups and product rounds. - Maths heavy?
Intuition + one worked example beats derivations for most roles. - Best 2-week revision?
Distributions → testing → one A/B case study, daily.
Read also: Interview Guide · ML Questions · DS Roadmap