Across the employers NACE surveys, an internship is frequently the start of a full-time job. NACE's 2026 Internship & Co-op Survey of 284 organizations reported an average intern conversion rate of 63.1% for 2024-25 interns, the highest in five years, and an acceptance rate of 88.3% among those interns. NACE's release doesn't break the figures out for data science. The questions in a data science internship interview deserve the preparation you'd give a job interview, because the average employer in NACE's sample converted 63.1% of its interns to full-time hires.
What do data science internship interviews actually test?
A data science internship interview tests foundations and judgment, at a level calibrated to coursework. Expect to be asked to write correct SQL, explain a statistical idea in plain words, spot a flawed evaluation, and describe your own project honestly. A production deployment on your resume is a bonus.
The field is worth the effort. The Bureau of Labor Statistics projects that employment of data scientists will grow 35 percent from 2025 to 2035, with about 24,800 openings a year, and puts the median annual wage at $120,230 in May 2025. BLS also notes that data scientists typically need at least a bachelor's degree, while some jobs require a master's or doctoral degree.
Data and ML roles are a small slice of Four-Leaf's job index. On October 21, 2026, they were 7.7% of all 81,101 role-tagged active postings.
Expect most of these rounds, in some mix. A SQL or Python screen. A statistics and probability round. An ML concepts round. A product or metrics case, sometimes with an A/B test inside it. A project walk-through. A behavioral conversation.
Which statistics questions come up for data science interns?
Data science intern statistics questions check whether you can explain a concept in plain words. The classic set is short. What is a p-value? What's the difference between a Type I and a Type II error? Why does the central limit theorem matter? How does a sample differ from a population? When is the mean a bad summary?
The p-value question is easy to fumble, because the textbook definition and the plain-language version drift apart under pressure. A p-value is the probability of seeing data at least as extreme as yours if the null hypothesis were true. It says nothing directly about whether the hypothesis is true. When the American Statistical Association released its statement on p-values in March 2016, its executive director, Ron Wasserstein, said that "the p-value was never intended to be a substitute for scientific reasoning."
The follow-up is the real question. If an interviewer asks what you would do with a p-value of 0.04, the strong answer asks about the effect size, the sample, how many tests were run and what decision depends on it. Interns who stop at "it's significant" have answered the easy half.
Practice each concept as a two-sentence explanation to someone outside the field, then as the follow-up you'd expect next.
Which SQL and Python questions should a data science intern expect?
SQL and Python questions for data science interns test whether you understand the shape of the data before you query it. Expect the grain of the data to decide the score more than the syntax does.
A typical SQL prompt asks for something like daily active users, the share of users who purchased within their first week, or the top product per category. Each one has a trap. Joining an orders table to an order-items table multiplies rows, so a COUNT(*) counts items rather than orders. A "first week" filter needs each user's signup date, which means a join or a window function rather than a WHERE clause on a calendar date. "Top product per category" needs ROW_NUMBER() or RANK() partitioned by category, so be ready to say which one you chose and what happens on a tie.
Python questions often use pandas. Handle missing values in a column, merge two frames and explain the row count that comes out, group and aggregate, then explain why merge produced more rows than you expected.
Say the grain out loud before you write anything. "One row per order" is a sentence that prevents most of the mistakes above. Four-Leaf's SQL interview guide and Python interview questions cover the full-time versions of both screens.
What machine learning questions do they ask interns?
Machine learning questions in an intern loop test evaluation judgment more than algorithm knowledge. You'll be asked to explain overfitting, compare precision and recall, describe what regularization does, or say why you'd pick a random forest over logistic regression. Those are fair, and you should be able to answer each in a minute.
Prepare hardest for the leakage question. The scikit-learn documentation defines data leakage as when "information that would not be available at prediction time is used when building the model," and states its general rule as "never call fit on the test data." An interviewer might describe a project where a scaler was fit on the full dataset before the train-test split and ask what went wrong. Or they might ask how you'd validate a model that predicts next month's churn, where a random split would let future rows into training.
Your own project is the best source of this answer. If you found a leak in your own work, or a metric that looked better than it should have, that story is worth more than any definition. If you never checked, check now, before an interviewer does.
For deeper ML preparation, Four-Leaf's guide on how to prepare for a data science interview covers the full-time loop.
How should you handle a product case or A/B test question as an intern?
A product case for a data science intern tests whether you can turn a vague question into a measurable one. Two prompts to prepare are "a metric dropped last week, what would you check" and "how would you test whether a new feature works."
For a metric drop, the strong answer checks the data before the business. Did logging change? Is the drop in one platform, region or app version? Did the definition of the metric change? Only then does it move to real causes like seasonality, a launch or a competitor.
For an A/B test, prepare four decisions. What is the unit of randomization? What is the primary metric, and what guardrail metric would stop the launch? How long will the test run, and why not stop it the day it looks significant? What would you do if the result is flat?
Know that peeking at a test and stopping early inflates false positives, and say so. Calculating sample size from memory matters less than knowing why it has to be fixed in advance.
What is overrated
Memorizing deep learning architectures is overrated for a data science internship. Prepare to explain why your model was evaluated the way it was before you study how a network is built.
Grinding hard algorithm problems is also overrated for analytics-leaning intern roles. One clean SQL query with the grain stated beats a hard algorithm solved in silence.
A long list of Kaggle medals is weaker than it looks. Competition data usually arrives with the problem framed and the test set held back by the organizers, so be ready to point to work where you made those choices yourself.
A one-week prep plan for a data science internship interview
- Write five SQL queries from scratch. A daily count, a first-week conversion, a top-N per group, a running total and a self-join. State the grain before each one.
- Explain six statistics concepts out loud. P-value, Type I and Type II error, the central limit theorem, confidence interval, sample versus population and correlation versus causation. Two sentences each, then the follow-up.
- Audit your best project for leakage. Find where the split happened and what was fit on what. Write down one thing you would change.
- Prepare one metric-drop walk-through. Data checks first, then causes, in under three minutes.
- Rehearse the project walk-through. Say which parts you built and which you didn't. Ownership is what project rounds probe, as Four-Leaf's guide to internship interview questions explains.
- Run the rounds out loud. Four-Leaf's AI mock interviews include a data scientist track with separate SQL and analytics, statistics, A/B testing, product metrics, ML modeling and behavioral rounds, each with follow-up questions.
Where this is heading
With NACE reporting a 63.1% average intern conversion rate across the employers it surveyed, a data science internship interview often carries the weight of a first job interview. The questions in this guide all come back to one skill, which is reasoning about data you haven't seen before.
Say how you would check a number before you report it. That habit carries into every round above.
