Data Analyst & Analytics Interview · Statistics & Experimentation · Lesson 1 of 3
Descriptive stats & distributions
Before any test, describe the data honestly. Central tendency and spread, plus the shape of the distribution, determine which summary and which test are even valid.
- Mean is sensitive to outliers; median is robust. Skewed data (revenue, latency) → report the median or a percentile, not the mean.
- Spread: variance / standard deviation; for skewed data prefer the IQR and percentiles (p50, p90, p99).
- Skew: right-skewed (long high tail) means mean > median. Most business metrics are right-skewed.
- Normal distribution: symmetric, ~68/95/99.7% within 1/2/3 SDs — the backbone of many tests.
Mean vs median in a barIf Bill Gates walks into a bar, the mean net worth of patrons skyrockets while the median barely moves. That gap is exactly why you report medians for skewed metrics like income, session length, or order value.
Central Limit TheoremThe CLT says the distribution of the sample mean approaches normal as n grows, even when the underlying data is skewed. That is why A/B tests on skewed metrics still use normal-based confidence intervals at scale.
◆ Lock it in
- Report the median/percentiles for skewed metrics; the mean misleads.
- Describe center + spread + shape before choosing a test.
- The CLT makes sample means approximately normal at scale, justifying standard CIs.
Feynman drill — say it out loudExplain why you'd report p90 latency instead of average latency, and what the CLT lets you assume about sample means.
Step 1 rate your confidence · Step 2 pick your answer
Order values are heavily right-skewed by a few whales. Which single number best represents a typical order?
Step 1 — how sure are you?