Say each answer aloud and swap the example for something you did. For questions that show up in every process, use the HR interview questions for freshers guide.
Contents
Technical/role knowledge
1. What is the difference between a dimension and a measure?
A dimension is the attribute you group or filter by, such as month, campus or product. A measure is the number you aggregate, such as a count or a sum. Say that swapping them is how a chart ends up with a bar for every row instead of a bar for every group.
Campus and month are dimensions. The number of applications is a measure. I group by campus and month, and I sum or count the measure.
2. When do you use an inner join, and when a left join?
An inner join keeps only rows that match in both tables. A left join keeps every row from the left table and fills the right side with nulls when nothing matches. Add the warning that a left join still duplicates the left row if the right table has several matches, which can double a sum.
I use an inner join for students who have an application. I use a left join when I also need students with none, and I check the grain before I sum so one student with two applications is not counted twice as a person.
3. What does a primary key protect you from in a report?
A primary key uniquely identifies a row, so the table does not store two rows for the same real-world thing. Duplicate keys, or a join that multiplies rows, make a sum count the same fact more than once. Say that you would check uniqueness before you trust a total.
If two rows share a student id, a count of rows is not a count of students. I would count distinct ids and compare that with a count of rows before I published the number.
4. When is the median a better summary than the mean?
The mean is the sum divided by the count, so a few very large values pull it. The median is the middle value once the numbers are ordered, and with an even count a common definition averages the two middle values. Use the median for skewed figures such as salaries or time taken, and say which one you chose.
If most people finish in a few days and two take months, the mean overstates a typical wait. I would report the median and mention the long cases separately rather than hide them inside the average.
5. What does GROUP BY do, and where do fresher queries go wrong?
GROUP BY collapses rows that share the same key into one row per group, and every other selected column has to be aggregated or included in the group. A common failure is selecting a column that is neither. Another is joining at the wrong grain first, so the aggregate double-counts.
Grouping applications by campus and summing the stipend is valid. Selecting the student name in that same query is not, unless I group by it too, because one campus has many names.
Problem solving and scenario
6. A stakeholder says the revenue number looks wrong. What do you check first?
Do not edit the chart. Confirm the definition, including what is included and which date the row uses, then the filters and the grain of the table. Compare the total with a known source for the same definition before you look for a visual bug.
I would ask whether the figure is invoiced or collected, and which month the date comes from. Then I would match that total to the source table. Only after that would I touch the chart.
7. A column has missing values. What do you decide before you fill them?
Count how many are missing and whether the gaps cluster, for example in one campus or one month. Ask what a blank means, because unknown is not the same as zero. Fill only when that meaning is agreed, and say in the note what you did.
If stipend is blank for one batch, I would not replace it with zero. Zero would look like unpaid. I would count the blanks, ask whether the batch was not on the scheme, and leave them out of the average until someone confirms.
8. Two dashboards show different totals for the same month. How do you reconcile them?
Put the two definitions side by side: filters, date boundaries, whether cancelled rows are included, and the grain after any join. Find the first rule that differs and recompute one total under the other rule. Do not average the two dashboards to settle the argument.
I would check whether one report uses the application date and the other uses the offer date. If that is the split, I would show both totals with the date named, rather than pick the larger one.
9. How would you define a conversion rate for event signups?
Write the numerator, the denominator and the time window before you name the rate. Both parts have to describe the same population, such as unique people who opened the form that week. Say the definition in the answer so two people cannot hear two different rates.
I would define it as signups divided by unique visitors to the registration page in that week. I would not divide by everyone in the college, because they were not all in a position to sign up.
10. Someone asks for a chart that would mislead. What do you do?
Name the distortion, such as an axis that does not start at zero or two scales that imply a relationship. Offer a chart or a table that keeps the comparison honest. Do not publish a view you cannot explain if someone reads the axis.
If a small change fills the chart because the axis starts at 90, I would say so and show the axis from zero, or show the table. I would not ship the cropped axis to make a trend look steeper.
HR and behavioural
11. Tell me about an analysis you did, even a college one.
State the question, the data, one cleaning choice and the limit of the result. Keep tools in a clause, not the whole answer. End with what you would check if you had another day.
I looked at which workshop slots filled up, using the signup sheet from our fest. I removed duplicate emails before I counted people. The sheet had no department for a third of the rows, so I did not claim a department pattern.
12. Describe a time your first conclusion was wrong.
Say what you believed, what check changed your mind, and what you do earlier now. The point is the check, not a dramatic failure. Do not invent a percentage improvement after you corrected it.
I thought attendance fell because of the time of day. A split by section showed one section had a test clash. I now break a drop by the groups I can name before I explain it with a single cause.
13. How do you explain a number to someone who does not work with data?
Use one sentence for the result, one for the comparison, and one for the caveat. Drop the query. Offer to show the rows behind the number if they want to argue with it.
Signups were higher than last week, and most of the increase was one department. The sheet still has duplicate emails in the older week, so I would clean that before I called it a trend.
14. Why data analysis, rather than another fresher role?
Connect it to a habit you already have: you like finding what a total is hiding. Mention a piece of work where the definition mattered more than the chart. Skip a generic claim that you love numbers.
I like the moment a total breaks into parts that tell different stories. In my project the interesting part was the missing rows, not the bar chart. I want a role where checking the definition is the job.
15. You have incomplete data and still need to recommend something. What do you say?
Separate what you observed, what you assumed, and what would change the recommendation. Give the smallest decision the evidence can support. Say which missing field you would collect before a bigger decision.
I would recommend we rerun the popular slot, because those rows are complete. I would not recommend dropping the other slot yet, because department is blank for many of those signups. That field is what I would collect next.