Data Analyst Interview Questions & Answers (2026)

Sample Answer

SELECT customer_id, SUM(order_amount) AS total_revenue FROM orders WHERE order_date >= CURRENT_DATE - INTERVAL '30 days' GROUP BY customer_id ORDER BY total_revenue DESC LIMIT 5. I would also add a JOIN to the customers table to pull readable names and include a HAVING clause if there's a minimum threshold.

Sample Answer

First I understand why data is missing — MCAR, MAR, or MNAR — because the mechanism affects the fix. For MCAR data I may impute with mean/median or drop rows if the volume is small. For MAR I use model-based imputation or multiple imputation. I never drop MNAR data without documenting the impact on results. I always check if the missingness itself is a signal worth analyzing.

Sample Answer

I start by defining the primary metric (conversion rate) and guardrail metrics (average order value, bounce rate). I calculate required sample size using a power analysis at 80% power and 5% significance. I randomly split users by user_id hash, not session, to avoid novelty effects. I run the test until sample size is reached, then use a z-test for proportions. I check for segment interactions before declaring a winner.

Sample Answer

An inner join returns only rows where the join condition is met in both tables. A left join returns all rows from the left table plus matching rows from the right — unmatched right-side columns are NULL. I use inner joins when I need clean, matched pairs; left joins when I want to preserve all records from one side and investigate what's unmatched.

Sample Answer

I lead with the business implication, not the methodology. Instead of 'our regression model shows a statistically significant p-value of 0.03,' I say 'customers who use the mobile app spend 22% more per order.' I use a single focused chart, limit slides to one insight per page, and anticipate the 'so what' and 'what should we do' questions before the meeting.

Sample Answer

I track MRR and ARR for revenue trajectory, churn rate and net revenue retention for retention health, customer acquisition cost (CAC) versus LTV for unit economics, and activation rate for onboarding effectiveness. I also watch the quick ratio (new MRR + expansion MRR) / churned MRR — anything above 4 indicates strong growth.

Sample Answer

I check row counts against expected volumes, verify primary key uniqueness, look for nulls in non-nullable fields, validate date ranges, and plot distributions for numeric columns to catch outliers and encoding errors. I cross-reference totals against a known source of truth, such as a financial report. I document every anomaly found and its resolution.

Sample Answer

Correlation means two variables move together statistically. Causation means one directly causes the other. Correlation can be spurious — ice cream sales and drowning rates are correlated because both increase in summer. To establish causation, you need a controlled experiment (RCT) or a quasi-experimental method like difference-in-differences, regression discontinuity, or instrumental variables.

Sample Answer

I score requests by business impact, urgency, and effort. I ask the requester 'what decision will this analysis inform?' If no decision is pending, I deprioritize. I batch similar queries and build reusable dashboards to eliminate repeat requests. I communicate timelines transparently and flag when an urgent request will delay a higher-value project.

Sample Answer

I started by interviewing the three stakeholders who would use it most to understand what decisions they made weekly. I prototyped in Google Sheets first to validate the metrics before investing in Tableau. I used their language, not data team jargon, for metric labels. I added a 'last updated' timestamp and a data freshness indicator. Six months later it was used in every weekly executive review.