Top 30 Google Data Analyst Interview Questions

1. How would you find the second-highest salary from an Employee table?

A common approach is to use DENSE_RANK():

SELECT salary
FROM (
    SELECT salary,
           DENSE_RANK() OVER (ORDER BY salary DESC) AS rnk
    FROM Employee
) t
WHERE rnk = 2;

Using DENSE_RANK() also handles duplicate salaries correctly.

2. What is the difference between INNER JOIN and LEFT JOIN?

• An INNER JOIN returns only rows that have matching records in both tables.
• A LEFT JOIN returns every row from the left table, even when there is no matching record in the right table.

A LEFT JOIN is useful when you need to identify records with no activity or matching data.

3. How would you find users who have no activity?

Use a LEFT JOIN and filter for NULL values in the joined table.

SELECT u.user_id
FROM Users u
LEFT JOIN Activity a
    ON u.user_id = a.user_id
WHERE a.user_id IS NULL;

The supplied research gives a similar example for finding neighborhoods with zero users.

4. How would you find the top 5 countries by total watch time?

Join the user and watch-history tables, aggregate watch time by country, then sort in descending order.

SELECT u.country,
       SUM(w.watch_minutes) AS total_minutes
FROM Users u
JOIN Watch_History w
    ON u.user_id = w.user_id
GROUP BY u.country
ORDER BY total_minutes DESC
LIMIT 5;

The research specifically includes this type of question as a reported interview example.

5. What is the difference between RANK(), DENSE_RANK(), and ROW_NUMBER()?

• ROW_NUMBER() gives every row a unique number.
• RANK() gives tied rows the same rank and leaves gaps.
• DENSE_RANK() gives tied rows the same rank without gaps.

For example, if two employees share second place:
RANK() → 1, 2, 2, 4
DENSE_RANK() → 1, 2, 2, 3

6. How would you find the top 3 salaries in each department?

Use a window function with PARTITION BY:

DENSE_RANK() OVER (
    PARTITION BY department_id
    ORDER BY salary DESC
)

Then filter for ranks less than or equal to 3.

7. What are window functions in SQL?

Window functions perform calculations across related rows while keeping the individual rows in the result. Common examples include:

• ROW_NUMBER()
• RANK()
• DENSE_RANK()
• LAG()
• LEAD()
• SUM() OVER()
• AVG() OVER()

They are useful for rankings, running totals, comparisons, and time-based analysis.

8. How would you calculate a running total?

Use SUM() with a window:

SUM(revenue) OVER (
    ORDER BY date
)

This calculates a cumulative total while keeping each date as a separate row.

9. How would you remove duplicate records in SQL?

One common approach is to use ROW_NUMBER():

ROW_NUMBER() OVER (
    PARTITION BY user_id
    ORDER BY created_at DESC
)

Keep the row where the generated number is 1. The research recommends this approach for deduplicating records while retaining the most recent entry.

10. What is the difference between GROUP BY and a window function?

• GROUP BY combines rows into aggregated results.
• A window function calculates across related rows without collapsing the original rows.

For example, GROUP BY department can return one row per department, while SUM(salary) OVER(PARTITION BY department) can show the department total alongside every employee.

11. How would you handle users with zero watch history?

Use a LEFT JOIN rather than an INNER JOIN. If the resulting value is NULL, use COALESCE() to treat it as zero:

COALESCE(SUM(w.watch_minutes), 0)

The research specifically highlights this as an important SQL edge case.

12. What should you do before writing a SQL query in an interview?

First clarify:

• What the question is asking.
• What each table represents.
• Which columns should be joined.
• How missing or duplicate records should be handled.
• What the expected output should look like.

Then explain your approach before writing the query. The research emphasizes clarifying assumptions and explaining your reasoning during technical problems.

13. What is a p-value?

A p-value represents the probability of observing data at least as extreme as the observed result, assuming the null hypothesis is true. A smaller p-value provides stronger evidence against the null hypothesis.

14. What is the difference between a t-test and a z-test?

• A t-test is generally used when the population standard deviation is unknown, particularly with smaller samples.
• A z-test is generally used when the population standard deviation is known or when large-sample assumptions apply.

The key is to explain why a particular test is appropriate rather than simply memorizing the definitions.

15. What is an A/B test?

An A/B test compares two versions of a product, feature, or experience. For example:

• Control: Existing version
• Treatment: New version

The analyst then compares a predefined metric to determine whether the observed difference provides evidence of an effect.

16. What is Sample Ratio Mismatch (SRM)?

Sample Ratio Mismatch occurs when the actual allocation of users between experiment groups differs significantly from the expected allocation. For example, an experiment planned as 50/50 might produce a substantially different split. The research notes that SRM can indicate problems such as randomization bugs, bots, or treatment failures.

17. What is Simpson's Paradox?

Simpson's Paradox occurs when a trend appears in multiple groups but changes or reverses when the groups are combined. This is why analysts should examine important segments and potential confounding variables rather than relying only on an overall average.

18. What would you do if an A/B test showed an unexpected result?

Start by checking:

• Data quality
• Experiment allocation
• Tracking or logging problems
• Sample size
• Statistical significance
• User segments
• External factors
• Possible experiment interference

Do not immediately conclude that the product change caused the result.

19. What is the difference between a Pandas Series and DataFrame?

• A Series is a one-dimensional labeled data structure.
• A DataFrame is a two-dimensional tabular structure made up of rows and columns.

For data analysis, DataFrames are commonly used to work with datasets.

20. What is the difference between loc and iloc in Pandas?

• loc is used for label-based indexing.
• iloc is used for integer-position-based indexing.

For example: df.loc[5, "salary"] selects using labels, while: df.iloc[5, 2] selects based on row and column positions.

21. How would you handle missing values in a dataset?

First understand why the values are missing. Depending on the situation, you might:

• Remove records
• Replace values using mean or median
• Use forward/backward filling
• Interpolate values
• Keep missing values if they carry meaning

The appropriate approach depends on the dataset and analytical objective.

22. How would you detect outliers?

Two common approaches are:

• Z-score: Useful when the data approximately follows a normal distribution.
• IQR method: Uses the interquartile range and is more robust for skewed data.

A common IQR rule flags values below: Q1 − 1.5 × IQR or above: Q3 + 1.5 × IQR

23. How would you choose the right chart for a dataset?

Choose the visualization based on what you want the reader to understand. For example:

• Bar chart: Compare categories
• Line chart: Show trends over time
• Heatmap: Show patterns across two dimensions

The goal is not to use the most complicated chart, but the chart that communicates the insight clearly.

24. Google Maps searches for “coffee shops” drop 15%. How would you investigate?

Use a structured approach:

• Check whether the data pipeline or tracking has an issue.
• Check recent product or app changes.
• Segment the decline by country, device, and user type.
• Check holidays, weather, or other external factors.
• Investigate possible market or competitor effects.

This exact type of Google Maps anomaly scenario appears in both supplied research documents.

25. How would you measure the success of a new Google Maps feature?

Start by defining what success means. Then consider several types of metrics:

• Adoption
• Engagement
• User success or utility
• Errors or frustration
• Performance
• Business or product guardrail metrics

For example, for an accessibility feature, you could examine feature adoption, successful route completion, route abandonment, and potential effects on app performance.

26. How would you solve a problem you have never seen before?

Start by clarifying the problem and identifying the constraints. Then:

• Break the problem into smaller parts.
• State your assumptions.
• Develop a possible approach.
• Test the approach using simple examples.
• Check edge cases.
• Explain the conclusion clearly.

The research emphasizes structured thinking and reasoning rather than simply jumping to an answer.

27. You have 8 balls and one is heavier. How can you find it in two weighings?

Divide the balls into groups of 3, 3, and 2. Weigh the two groups of three:

• If they balance, the heavier ball is among the remaining two.
• If they do not balance, it is in the heavier group.

Then use the second weighing to identify the heavier ball within the remaining candidates. The supplied research presents this as a divide-and-conquer logic problem.

28. How would you approach a difficult SQL or analytics problem if you get stuck?

Do not immediately start guessing. Explain what you understand, identify what is unclear, and break the problem into smaller steps. You can also test a simpler version of the problem first. The research specifically recommends clarifying the problem, outlining the approach, checking intermediate results, and communicating assumptions.

29. Tell me about a time your analysis proved your initial assumption was wrong.

Use the STAR framework:

• Situation: Explain the original problem.
• Task: Describe what you needed to determine.
• Action: Explain how you analyzed the data and discovered the assumption was incorrect.
• Result: Explain what changed and what you learned.

The important part is demonstrating that you can change your conclusion when the evidence does not support your original assumption.

30. How would you explain a complex analytical concept to a non-technical stakeholder?

Avoid unnecessary technical terminology. Start with the business problem, explain the concept using a simple example or analogy, and then connect it to the decision that needs to be made. For example, instead of explaining a complicated data model technically, you could compare it with categories in a household budget to make the relationship easier to understand.