Data analyst interviews commonly test more than whether you can write a query: expect to show how you check data, interpret results, choose tools, and explain what a finding means for a decision. The 50 prompts below are representative practice questions, not a frequency-ranked list or a prediction of what any specific employer will ask. Interview formats and tool requirements vary, so tailor your practice to the job description.
For open-ended questions, a useful approach is to clarify the goal and metric, check definitions and data quality, state assumptions, choose an analysis, validate the result, explain uncertainty, and connect the finding to an action. For experience questions, use your own examples rather than presenting a hypothetical as work you have done.
How to use these 50 questions
Work through the sections most relevant to the role. A job centered on reporting may emphasize spreadsheets, dashboards, and stakeholder communication; a product or experimentation role may probe SQL, metrics, and statistical reasoning. Google’s Data Analytics Certificate page describes the work as preparing, processing, and analyzing data, sharing findings with stakeholders, and making data-driven recommendations. Its curriculum includes spreadsheets, SQL, presentation tools, Tableau, Python, and Kaggle, but an employer’s job description is the better guide to what tools to prioritize.
Interview guides describe different possible formats, including technical exercises, case discussions, and other interview stages; there is no single process to assume. One practice guide explicitly says its original questions are not actual or leaked interview questions. Treat the prompts here as preparation, not a forecast.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →#1 Best Overall
SQL and data retrieval
1. What is the difference between an INNER JOIN and a LEFT JOIN?
An INNER JOIN returns rows with matching keys in both tables. A LEFT JOIN keeps every row from the left table and adds matching values from the right; unmatched right-side values are NULL. I would choose based on whether unmatched left-side records should remain in the result, and check whether keys are unique to avoid unexpectedly multiplying rows.
2. How would you find total sales by month?
I would confirm which date defines a sale and which statuses count, then group valid transactions by a year-month date bucket and sum the sales amount. I would check the currency, treatment of refunds, and whether the output includes months with no sales.
3. How would you find customers who have never placed an order?
I would start from the customer table and use a LEFT JOIN to orders, then filter for rows with no matching order key. I would verify that the join uses the correct customer identifier and that cancelled or test orders are treated according to the business definition.
4. What is the difference between WHERE and HAVING?
WHERE filters input rows before grouping; HAVING filters groups after aggregation. For example, I would use WHERE to limit orders to a date range and HAVING to retain only customers whose resulting order count exceeds a threshold.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →5. How would you calculate the average order value by customer?
I would first clarify whether the metric is average transaction value per order or average revenue per customer, since those are different calculations. For order value, I would aggregate eligible line items to order level if needed, then average those order totals by customer. I would state how refunds, discounts, and incomplete orders are handled.
6. When would you use a CTE or subquery?
Either can express an intermediate result. I use a CTE when naming stages makes a multi-step query easier to read or validate; a subquery can be compact for a simple transformation. I would also consider the database engine’s behavior for performance rather than assuming a CTE is always faster.
7. How do window functions differ from GROUP BY?
GROUP BY collapses rows into groups, while a window function calculates across related rows without removing the individual rows. I might use a window function to rank purchases within each customer or calculate a running total while retaining each transaction.
8. How would you rank products by revenue within each category?
I would aggregate revenue by category and product, then apply a ranking window function partitioned by category and ordered by revenue descending. I would clarify whether ties should share a rank and whether zero-revenue products belong in the output.
Free tools Windows power users keep installed
One-click scans. No signup required.
9. A query returns more rows after a join than before. How would you investigate?
I would compare row counts and key uniqueness on each side, then inspect the join keys for one-to-many or many-to-many matches. I would test a small set of affected keys and confirm whether the extra rows are legitimate or indicate a wrong key, duplicate records, or a grain mismatch.
10. How do you make a SQL analysis reproducible and easier to review?
I would use clear aliases, consistent definitions, explicit filters, and named intermediate steps where helpful. I would document assumptions that affect the result and validate key totals against an independent query or trusted report before sharing it.
Rank #2
- Ideal for Gifting
- Ideal for a bookworm
- Compact for travelling
Statistics and experimentation
11. What do mean, median, and mode tell you?
They summarize different aspects of a distribution. The mean is sensitive to extreme values, the median is the midpoint and is often more representative of skewed data, and the mode is the most common value or category. I would choose based on the distribution and the question rather than reporting one automatically.
12. What is standard deviation?
Standard deviation describes the typical spread of values around the mean in the units of the measured variable. I would interpret it alongside the distribution and context; the same standard deviation can mean different things for different scales or highly skewed data.
13. What does a p-value mean?
A p-value is the probability, assuming the null hypothesis and the statistical model are true, of observing results at least as incompatible with the null as the observed result. It is not the probability that the null hypothesis is true, nor does it measure practical importance.
14. What is a confidence interval?
A confidence interval is a range produced by a method that, over repeated samples under its assumptions, would cover the true parameter at the stated rate. I would use it to communicate estimate uncertainty, while noting that a narrow interval does not remove bias or poor measurement.
15. How would you distinguish correlation from causation?
Correlation describes an association; it does not establish that one variable caused the other. Confounding, reverse causality, or selection effects may explain the pattern. To support a causal claim, I would look for a suitable design, such as a randomized experiment, or a defensible observational method with stated assumptions.
16. How would you design an A/B test?
I would define the decision, primary outcome, eligible population, and randomization unit first. Then I would set guardrail metrics, estimate the sample and duration needed, ensure assignment is implemented correctly, and plan how to handle exclusions and multiple comparisons. After the test, I would check data quality and interpret both uncertainty and practical impact.
17. An experiment is statistically significant but the effect is small. What would you tell the team?
I would report the estimated effect and its uncertainty, then compare the size with the decision threshold, implementation cost, and possible harms. Statistical significance alone does not establish that an effect is important enough to act on.
18. What is selection bias, and how could it affect an analysis?
Selection bias occurs when inclusion in the data or analysis is related to the factors being studied, producing a sample that does not support the intended inference. I would examine who is missing or excluded, compare included and excluded groups where possible, and limit conclusions to the population the data can represent.
Spreadsheets and analytical tools
19. How would you use a pivot table to summarize transactions?
I would place the dimension of interest, such as region or month, in rows or columns and the measure, such as transaction amount, in values. I would confirm the aggregation type, filter out ineligible records, and compare the pivot total with a source total.
20. When would you use a lookup function?
I would use a lookup to bring an attribute from a reference table into a working dataset, such as a product category keyed by product ID. Before relying on it, I would check key uniqueness, unmatched keys, and whether the lookup is exact rather than an unintended approximate match.
Recommended Free Tools
21. How would you find duplicate records in a spreadsheet?
I would first define what constitutes a duplicate: an identical row, repeated transaction ID, or repeated combination of fields. Then I would use a duplicate check or count by key, inspect examples, and avoid deleting records until I know whether repeats reflect valid events or data errors.
22. How would you decide between a spreadsheet, SQL, and Python?
I would consider data size, repeatability, collaboration needs, complexity, and the tools available in the role. A spreadsheet can be effective for a small, transparent analysis; SQL is useful for retrieving and aggregating database data; Python can support repeatable transformations or more complex analysis. The choice should fit the task and the team’s workflow.
23. How would you make a spreadsheet analysis auditable?
I would keep raw inputs separate from calculations and outputs, label assumptions, use consistent formulas, and avoid unexplained hard-coded values. I would test formulas on known cases and include checks that reconcile totals or flag missing inputs.
Python and analytical workflow
24. How would you group and summarize data in pandas?
I would select the relevant columns, confirm types and missing-value handling, then use grouping and aggregation to produce the required grain. I would inspect the resulting row count and compare totals with the input or a separate calculation.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minute25. How would you merge two data frames?
I would identify the intended key and relationship, check key uniqueness, and choose the join type based on which records need to be retained. After merging, I would inspect unmatched records and row counts to detect duplication or dropped data.
26. How would you handle missing values in Python?
I would first quantify missingness by field and relevant segment, then investigate whether values are missing randomly, by process, or because of an instrumentation issue. Depending on the analysis, I might exclude records, impute values, or retain a missing category, but I would state the choice and test how it affects the conclusion.
27. How do you validate a data transformation?
I would define expected properties before transforming, such as row counts, unique keys, valid ranges, and reconciled totals. I would test edge cases, compare before-and-after summaries, and keep the transformation steps reproducible so errors can be traced.
28. What does reproducibility mean in an analysis?
Another analyst should be able to understand and rerun the steps using the same inputs and definitions, and obtain the same result. I would preserve the code or formulas, note data sources and assumptions, and avoid hidden manual edits that cannot be recreated.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteData quality and metrics
29. How would you investigate missing values?
I would measure how much is missing, identify where and when it occurs, and compare missingness across relevant groups. I would then look for a collection or pipeline cause and decide whether the field can support the intended analysis without an explicit caveat or alternative method.
30. What would you do with an outlier?
I would verify whether it is a data-entry, unit, or processing error, then determine whether it is a valid extreme observation. I would not remove it solely because it is inconvenient; I would assess its influence and, if useful, show results with and without a justified treatment.
Rank #4
31. How would you detect a tracking or instrumentation gap?
I would compare event volumes and key properties over time, across platforms, and against independent business signals where available. A sudden break in event patterns could reflect a real behavior change or a logging change, so I would check release and pipeline history before interpreting it as user behavior.
32. How would you define a conversion rate?
I would specify the qualifying conversion event, eligible population, time window, and denominator. For example, the rate could be users completing an action divided by users who reached a defined prior step during a stated period. I would document exclusions and ensure numerator and denominator use compatible units.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
33. What is the difference between a leading and a lagging metric?
A leading metric can provide an earlier signal about a future outcome, while a lagging metric records an outcome after it occurs. I would treat a leading metric as a proxy, not a guaranteed predictor, and validate whether changes in it reliably align with the outcome of interest.
Visualization and communication
34. How would you choose a chart for a dataset?
I would start with the comparison: use a line chart for change over time, bars for category comparisons, and a scatterplot for the relationship between two numeric variables. I would keep scales and labels clear, avoid unnecessary decoration, and make the chart answer one question without hiding relevant context.
35. What makes a dashboard useful?
A useful dashboard is built around a specific audience and decision. It should have clearly defined metrics, an appropriate level of detail, readable labels, sensible filters, and visible dates or refresh context. I would verify that users can distinguish a real change from data delay or a definition change.
36. How would you explain an uncertain result to a nontechnical audience?
I would lead with the decision-relevant estimate, explain the range of plausible outcomes in plain language, and say what the data cannot establish. I would avoid presenting uncertainty as a technical footnote if it changes the recommended action.
37. How would you present a finding that contradicts what stakeholders expected?
I would show the evidence and definitions, explain the checks I performed, and distinguish observed results from interpretation. I would invite review of assumptions or missing context without changing the result to fit expectations, then outline what additional evidence could resolve remaining uncertainty.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Business cases and diagnostic questions
38. How would you investigate a 10% drop in daily views?
I would first confirm the metric definition, comparison period, and whether the drop is outside normal variation. Then I would check tracking and pipeline health, segment by platform, geography, acquisition source, and user cohort, and look for a release, seasonality, or traffic-mix change. I would identify the segments driving the decline, estimate impact on downstream outcomes, and recommend a next step based on the evidence.
39. Revenue is up but the number of orders is down. What might explain this?
I would decompose revenue into order volume and average order value, then examine price, product mix, discounting, and customer segments. I would check refunds and timing effects as well as whether the two metrics use matching definitions and periods before attributing the change to customer behavior.
40. A product team asks, “Did the new feature work?” What do you ask first?
I would clarify the intended outcome, target users, launch timing, and what success or harm would look like. I would ask whether there is an experiment or comparison group, then define primary and guardrail metrics before analyzing results.
Best Value
- It can be a gift option
- Comes with secure packaging
- Helpful in various ways
41. How would you decide whether to prioritize a metric decline?
I would assess the magnitude, duration, affected population, data reliability, and likely consequence for the business or customers. I would compare the decline with normal variability and other indicators, then recommend investigation or action proportionate to the risk.
42. A stakeholder asks for a chart but has not explained the decision. What would you do?
I would ask what they need to decide, who will use the chart, and what comparison would change their view. Then I would choose the measure and visual that support that decision, rather than producing a chart whose relevance is unclear.
Behavioral and project questions
43. Tell me about a data project you worked on.
Structure the answer around the problem, your responsibility, the data and methods, the checks you performed, the finding, and the decision or outcome. Be precise about your own contribution and explain limitations rather than claiming the analysis proved more than it did.
44. Tell me about a time your analysis influenced a decision.
Describe the decision-maker’s question, the evidence you produced, how you communicated uncertainty, and what decision followed. If the recommendation was not adopted, explain what you learned or what further evidence was needed.
Recommended Free Tools
45. Tell me about a mistake you made in an analysis.
Choose a real example, explain how the mistake occurred and how you detected it, then describe the correction and any process change that reduced the chance of recurrence. Emphasize accountability and validation rather than implying that mistakes never happen.
46. How have you handled an ambiguous request?
Explain how you clarified the objective, identified assumptions, proposed a tractable first analysis, and checked that the result answered the stakeholder’s actual need. A strong answer shows progress without pretending ambiguity was fully resolved at the outset.
47. Describe a disagreement with a stakeholder or colleague about data.
Use a real example and focus on how you compared definitions, evidence, and assumptions. Explain how you reached a decision or surfaced an unresolved issue respectfully, without portraying disagreement as a contest to win.
48. How do you prioritize competing analysis requests?
I would compare urgency, expected impact, decision deadlines, effort, and dependencies, then make trade-offs visible to requesters. If priorities conflict, I would seek alignment with the relevant lead rather than silently delaying work or overpromising.
49. What do you do when you do not know the answer to a technical question?
I would be transparent about what I do not know, then reason from the goal and available information. I might clarify assumptions, outline how I would investigate, and identify what I would validate before giving a firm answer.
50. Why do you want this data analyst role?
Connect your genuine interest and relevant experience to the role’s domain, responsibilities, and tools. Give a specific reason grounded in the job description, and explain how your strengths could help the team make better decisions.
Am I ready for the interview?
You are better prepared when you can explain not only an answer, but also why your method fits the question and what could make the conclusion unreliable. Use this checklist to find gaps:
- You can write and explain the SQL patterns named in the role, including joins and aggregation.
- You can interpret uncertainty and distinguish an association from a causal claim.
- You can describe how you check missing values, duplicates, outliers, definitions, and tracking.
- You can choose an appropriate tool and explain the trade-off for the task.
- You can turn an ambiguous business change into a structured investigation and recommendation.
- You have truthful, specific examples of your projects, decisions, mistakes, and collaboration.
Google estimates that its Data Analytics Certificate can be completed in under six months with fewer than ten hours of flexible study each week. That is an estimate for that program, not a timetable for preparing for a particular interview.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
