48 lessons
Statistics · Module 2Module material

EDA, Evidence and Misleading Statistics interview questions

28 questions interviewers ask on this module’s topics, from first principles to the follow-ups. Each comes with a model answer.

Exploratory data analysis

  1. What is exploratory data analysis, and what should come first?●
  2. What is a unit of analysis, and why does choosing the wrong one matter?●
  3. A column has missing values and the table has exact duplicate rows. Should you delete them before the analysis?●●
  4. Why reindex a daily series to the full calendar?●
  5. What does a zero in a transaction-based daily count tell you?●●

Charts and presentation

  1. Why should a bar chart's axis start at zero?●
  2. A dashboard shows two weeks and claims sustained growth. What do you ask for?●●
  3. How can smoothing or aggregation hide a change?●●
  4. Why are dual-axis charts risky?●

Denominators and selection

  1. What is a base rate?●
  2. A lender reports a low default rate among approved clients. What can it say about all applicants?●●
  3. What is wrong with filling unobserved outcomes with zero?●●
  4. How does selection bias affect an ML model trained on approved applicants?●●

Subgroups and Simpson's paradox

  1. What is Simpson's paradox?●
  2. Why did the Berkeley aggregate reverse?●●
  3. What is standardization, and what question does it answer?●●
  4. Does the Berkeley table show whether admissions were fair?●●

Selective reporting and survivorship

  1. What is selective reporting?●
  2. What is the winner's curse, and how large is it here?●●
  3. What is survivorship, and how do you guard against it?●●
  4. Why is early stopping a form of selection?●

Claims, prediction and causation

  1. A metric rose after a feature launch. Can you attribute the rise to the launch?●●
  2. Why randomize customers rather than invoices?●●
  3. A randomized simulation gives a difference of 0.0221. What else belongs in the report?●●
  4. What is the difference between a prediction claim and a causal claim?●

EDA for an ML decision

  1. Why seal a holdout before EDA for a model?●●
  2. A table has no missing cells. Is it clean?●●
  3. What belongs in an EDA report for a later model?●●

Directory listings

  • SchoolWhool on Siteefy