Data Analyst Interview Questions Quiz: 10 SQL and Scenario Questions

Data Analyst Interview Questions Quiz · jobquix.sbs

All questions

1. A query LEFT JOINs customers to orders on customer_id, groups by customer, and selects COUNT(o.order_id). What does it return for a customer with no orders?
   • A. 0, because COUNT on a column skips the NULL that the LEFT JOIN creates
   • B. 1, because the LEFT JOIN still produces one row for that customer
   • C. NULL, because there are no matching orders to count
   • D. Nothing, because customers with no orders are dropped by the join before the GROUP BY runs
   Show the best answer

   A. A LEFT JOIN keeps every customer and fills the order columns with NULL when there's no match. COUNT(column) ignores NULLs, so the result is 0. COUNT(*) would count the row itself and return 1, which is a common interview trap.

2. You LEFT JOIN customers to orders, then add WHERE o.order_date >= '2026-01-01'. Customers with no 2026 orders disappear from the results. Why?
   • A. LEFT JOIN only keeps unmatched rows when the query has no WHERE clause at all, so any filter will drop customers without orders no matter where it's placed.
   • B. The date needs to be cast to a timestamp before it can be compared to a string.
   • C. GROUP BY removes rows with NULL values, so those customers were filtered out during aggregation instead.
   • D. The WHERE filter rejects the NULL order dates, so it acts like an inner join. Move the condition into the ON clause.
   Show the best answer

   D. For unmatched customers, o.order_date is NULL, and a comparison with NULL is never true, so WHERE removes those rows. Putting the date condition in the ON clause filters orders before the join and keeps every customer. Interviewers ask this to see whether you understand how joins and filters interact.

3. You need total sales by region, but only for regions with more than $50,000 in sales. Where does the $50,000 condition belong?
   • A. In the WHERE clause, as WHERE sales > 50000
   • B. In a HAVING clause after GROUP BY, as HAVING SUM(sales) > 50000
   • C. In the ORDER BY clause, so the largest regions come first and you can stop reading at $50,000
   • D. In a subquery that filters out any individual sale under $50,000 before grouping
   Show the best answer

   B. WHERE filters individual rows before grouping, while HAVING filters groups after aggregation, so a condition on SUM(sales) goes in HAVING. WHERE sales > 50000 or a subquery on single sales would keep only large individual transactions, which answers a different question.

4. You join orders to order_items (one order has many items), then SUM(orders.shipping_fee). The total is much higher than finance reports. What is the most likely cause?
   • A. SUM ignores NULL shipping fees, which pushes the total up.
   • B. Finance is probably using a different currency conversion for the same period.
   • C. The join repeats each order's shipping fee once per item, so the fee is counted multiple times.
   • D. The query needs an ORDER BY so the database adds the rows in the correct sequence, otherwise the sum can drift from run to run.
   Show the best answer

   C. Joining across a one-to-many relationship repeats the "one" side, so order-level values get summed once per item. Sum shipping fees from the orders table alone, or aggregate items to the order level before joining. Checking row counts before and after a join catches this quickly.

5. About 12 percent of rows in a customer survey are missing the income field. What is the best first step?
   • A. Fill the blanks with 0 so every row can be used in the analysis.
   • B. Delete every row with a missing value, since incomplete data is unreliable and keeping it could bias the averages you report.
   • C. Fill the blanks with the overall average income so the dataset is complete and the mean stays the same.
   • D. Check whether those rows differ from the rest, such as by age or channel, then choose a method and document it.
   Show the best answer

   D. How you handle missing data depends on why it's missing. If certain groups skip the question more often, dropping or averaging can bias your results. Filling with 0 is almost always wrong for a field like income, and any choice you make should be documented.

6. A sales director asks why a dashboard shows revenue down 8 percent. How should you present your analysis?
   • A. Lead with the main finding and its business impact in plain language, show one clear chart, and keep method details ready if they ask.
   • B. Walk through your SQL query and data cleaning steps first so they trust how the number was produced before they see it.
   • C. Send the full spreadsheet so they can explore the numbers and reach their own conclusions.
   • D. Explain the statistical tests and confidence intervals you used, because precise language prevents misunderstandings later and shows the analysis is rigorous.
   Show the best answer

   A. Non-technical stakeholders need the answer and what it means for their decisions first. Method details build trust when someone asks, but leading with them buries the finding, and a raw spreadsheet hands the analysis back to the person who asked you to do it.

7. "How do you design a dashboard for a busy executive?" Which answer is strongest?
   • A. I include every metric the team tracks so the executive never has to ask for more data.
   • B. I use a lot of chart types and colors so the dashboard looks impressive and holds attention in meetings, and I add animations when the tool supports them.
   • C. I start by asking what decisions they make, then show a few key metrics with targets and trends at the top and details below.
   • D. I copy the layout of the last dashboard I built so everything looks consistent across the company.
   Show the best answer

   C. Good dashboards are built around the user's decisions, with the most important metrics and their context, like targets and trends, easy to scan. Too many metrics or decorative charts make the key numbers harder to find, and reusing a layout ignores what this audience needs.

8. An A/B test of a new checkout button shows a 2 percent lift in conversion with a p-value of 0.03, using a 0.05 significance level. What is the most accurate interpretation?
   • A. There is a 97 percent chance the new button is better, so the team can expect the 2 percent lift to continue after launch.
   • B. The result is statistically significant: a difference this large would be unlikely if the buttons truly performed the same.
   • C. The lift is guaranteed to hold at about 2 percent once the button rolls out to every user, since the result passed the significance threshold.
   • D. The test proves nothing, because the lift is under 5 percent.
   Show the best answer

   B. A p-value below the chosen level means the observed difference would be unlikely if there were no real effect. It isn't the probability that the variant is better, and it doesn't guarantee the size of the lift. A strong answer also checks practical impact and whether the test ran its planned length.

9. Before sharing a weekly report built on a new data source, which check matters most?
   • A. Make sure the chart colors match the company's brand guidelines.
   • B. Confirm the query runs in under a minute so the report refreshes quickly every morning without timing out.
   • C. Check that the report has more rows than last week's version.
   • D. Reconcile key totals against a trusted source and check for duplicates, nulls and unexpected values.
   Show the best answer

   D. Reconciling totals with a trusted source, such as a finance report, and checking for duplicates, nulls and out-of-range values catches most errors before anyone acts on bad numbers. Speed and formatting matter only after the numbers are right, and more rows can mean duplicates, not better data.

10. You need to pull 40 million transaction rows, clean them and run the same analysis every month. Which tool choice makes the most sense?
    • A. Use SQL to filter and aggregate in the database, then Python for cleaning and a repeatable script.
    • B. Do it all in Excel, since it's what most stakeholders already know and they can check your formulas themselves.
    • C. Export everything to CSV and clean it by hand each month, so you can see every change you make to the data and catch problems before they reach the final numbers.
    • D. Use whichever tool you learned first, since all three handle any size of data the same way.
    Show the best answer

    A. SQL is built to filter and aggregate large tables where they live, and Python makes cleaning and repeat analysis scriptable. A single Excel worksheet holds just over 1 million rows, and manual monthly cleaning is slow and hard to reproduce. Excel still works well for small, one-off analyses and sharing results.