Technical interview guide

Regression Analysis

Modeling a relationship between variables — linear regression's assumptions, and what R² does and doesn't tell you.

Read: 46 min
Practice MCQs: 25
Interview QA: 25
Edition: v6
Editorial status: Reviewed
Relevant for: AI EngineerBusiness Intelligence DeveloperComputational BiologistData AnalystData ScientistMachine Learning EngineerProduct ManagerQuantum Software Engineer

Scope: Standards and library references current as of 2026-09: NIST/SEMATECH e-Handbook, statsmodels, scikit-learn, and SciPy.

Interview QA

Treat each question like a live interview question: answer out loud first (structure, assumptions, tradeoffs), then open the model answer to spot gaps and rehearse a tighter follow-up.

Curated: 2026-10-05 · Written: 2026-10-05 · Reviewed: 2026-10-03

QA-1

Explain what least squares actually fits and how it changes a production decision.

QA-2

How would you reason about interpreting a coefficient in a system you own?

QA-3

Walk through linearity is about parameters, including where engineers most often get it wrong.

QA-4

What does the residual plot guarantee, and what does it deliberately leave open?

QA-5

Describe heteroskedasticity and the evidence you would collect before relying on it.

QA-6

How do you detect autocorrelation and heteroskedasticity in regression residuals, and what are their consequences for OLS inference?

QA-7

Where does multicollinearity matter, and where is it irrelevant?

QA-8

Teach omitted variable bias to an engineer who has only seen it as a rule of thumb.

QA-9

What is collider bias, how does conditioning on a collider induce spurious association, and how does it differ from confounding?

QA-10

What are the limitations of R-squared as a measure of model fit, and how does adjusted R-squared address them?

QA-11

What would you measure before treating prediction intervals versus confidence intervals as settled?

QA-12

How does extrapolation change if the workload grows by two orders of magnitude?

QA-13

How do you distinguish outliers, high leverage, and influential observations, and what diagnostic metrics identify them?

QA-14

When would you refuse a design because of categorical predictors?

QA-15

How would you explain interactions without using the usual slogan?

QA-16

What failure would you inject to check a team's understanding of regularisation?

QA-17

How should on-call treat an alert that names scaling before regularisation as the cause?

QA-18

What belongs in a runbook section on logistic regression, and what does not?

QA-19

How would you review a pull request whose risk is separation in logistic models?

QA-20

What trade-off does count outcomes force that a junior answer usually skips?

QA-21

How would you brief product on why transforming the outcome delays a ship date?

QA-22

What is the smallest experiment that would change your mind about missing data?

QA-23

How does regression to the mean interact with rollback, and where do people ignore that?

QA-24

What would you ask a candidate who recites model selection and validation but cannot apply it?

QA-25