Data Analyst Interview Questions and Answers

Questions and model answers

9 questions you are likely to hear.

1. 1

   What is the difference between INNER JOIN and LEFT JOIN, and when would a LEFT JOIN give you wrong totals?

   Why they ask: Join logic is the most common source of analyst errors, so interviewers test both the definition and the pitfalls.

   Sample answer: An INNER JOIN returns only rows with a match in both tables, while a LEFT JOIN keeps every row from the left table and fills unmatched columns with nulls. A LEFT JOIN can inflate totals when the right table has multiple rows per key. For example, joining orders to a payments table where one order has two partial payments duplicates the order amount. I always check the grain first and either aggregate the right table in a CTE before joining or count distinct keys after the join. In one audit this check found revenue overstated by around 8 percent on a sales dashboard.

2. 2

   Write a query to find the second highest salary in each department.

   Why they ask: This checks window function knowledge and how you handle ties.

   Sample answer: I would use DENSE_RANK so ties are handled sensibly. In a CTE I select department_id, employee_name, salary and DENSE_RANK() OVER (PARTITION BY department_id ORDER BY salary DESC) as rnk, then select from the CTE where rnk equals 2. DENSE_RANK means if two people share the top salary, the next distinct salary is still ranked 2. If the business wanted exactly one person, I would use ROW_NUMBER with a tie-breaker such as hire date. I would also confirm with the interviewer whether departments with only one employee should appear with a null or be excluded.

3. 3

   How do you handle missing values in a dataset?

   Why they ask: Interviewers want judgement, not a single rule like always drop nulls.

   Sample answer: First I find out why the data is missing, because the fix depends on the cause. If a field is missing at random and the share is small, dropping those rows may be fine. If it is systematic, such as delivery date missing for all cancelled orders, it carries meaning and I keep it as its own category. For numeric fields I might impute the median, not the mean, when the distribution is skewed. In a customer analysis in pandas I found 15 percent of ages missing for one acquisition channel, which turned out to be a form bug we reported to the product team.

4. 4

   Our conversion rate dropped 10 percent last week. How would you investigate?

   Why they ask: This classic case question tests structured thinking and business sense.

   Sample answer: I would first confirm the drop is real by checking tracking changes, data pipeline delays and whether the definition changed. Then I segment: by device, browser, traffic source, country, new versus returning users and funnel step. Usually the drop concentrates in one segment. I would also check releases, campaigns and pricing changes in that week. In a similar case at an ecommerce startup, the fall was entirely on Android at the payment step, which matched an app release that broke a UPI payment option. Rolling back recovered conversion within two days.

5. 5

   Explain a p-value to a non-technical manager.

   Why they ask: This tests both statistical understanding and communication.

   Sample answer: I would say: a p-value tells us how surprising our result would be if the change we made actually had no effect. If the p-value is 0.03, there is about a 3 percent chance of seeing a difference this large just by random luck. We usually treat anything under 0.05 as strong enough to act on. I would add two cautions: a small p-value does not tell us the effect is big enough to matter for the business, and if we test twenty things at once, one will look significant by chance, so we plan tests carefully before running them.

6. 6

   How would you design a dashboard for a sales leadership team?

   Why they ask: The interviewer checks whether you start from decisions and users rather than from charts.

   Sample answer: I start by asking what decisions the team makes weekly and which three or four numbers drive them, usually revenue against target, pipeline coverage, win rate and average deal size. The top row shows those KPIs with comparison to target and last period. Below that I add a trend line and a breakdown by region and rep, with filters for time and product. I avoid more than six visuals per page. For a regional sales team I built this in Power BI with drill-through to account level, and weekly review meetings shortened because people stopped asking for separate Excel extracts.

7. 7

   What is the difference between correlation and causation? Give an example from work.

   Why they ask: Interviewers want to know you will not mislead stakeholders with spurious conclusions.

   Sample answer: Correlation means two variables move together; causation means one actually drives the other. At a subscription company we saw that customers who used the mobile app had 40 percent lower churn, and marketing wanted to push everyone to the app. I pointed out that engaged customers are more likely to install the app in the first place. We ran a controlled experiment that prompted a random half of web users to install the app. The real churn effect was much smaller than the raw correlation suggested, which saved the team from over-investing in that campaign.

8. 8

   How do you make sure your numbers are accurate before you share them?

   Why they ask: This tests rigour and ownership of data quality.

   Sample answer: I have a short checklist. I reconcile totals against a trusted source, such as finance revenue for the month, and I check row counts before and after each join. I look at the distribution of key fields for impossible values like negative quantities or dates in the future. I compare the result with last period to see whether the change is plausible. For important reports I ask a colleague to review the logic. Once, this caught a timezone issue where UTC timestamps were shifting late-night Indian orders into the next day.

9. 9

   Tell me about an insight you found that changed a decision.

   Why they ask: This is the impact question: it separates report builders from analysts.

   Sample answer: At a food delivery client, management planned to expand discounts across all cities to boost orders. I ran a cohort analysis in SQL and Python and found that discount-acquired customers in two tier-2 cities had a 30-day retention rate less than half of organically acquired users, while in metros the gap was small. I presented one chart comparing retention curves with a cost-per-retained-customer estimate. Leadership limited the discount to metros and redirected the remaining budget to delivery speed in the tier-2 cities, which was the main complaint in reviews.

Keep going.

• SQL Developer Interview Questions and Answers
• Power BI Developer Interview Questions and Answers
• Data Scientist Interview Questions and Answers
• SQL Job Support Online for Developers and Analysts
• Auto Apply to Data Analyst Jobs with AI
• AI mock interviews