LOR Sample for MS in Data Science: What to Write

LOR Sample for MS in Data Science: What to Write

A recommendation letter for an MS in Data Science should prove one thing a transcript cannot: that the student can reason about data. Rice's online Master of Data Science asks for two letters from people familiar with the applicant's "interest in data science and analytics". UChicago's MS in Applied Data Science asks for two and strongly recommends one from a workplace supervisor.

A template letter that lists languages and libraries tells a committee nothing the CV does not. This page shows what three data science programs actually ask recommenders for, then gives a constructed letter built around the evidence they ask for: a flawed analysis the student caught. If you are the recommender and the draft is ready, choose the graduate school recommendation letter rubric for a free Quick Review.

Every program page quoted below was checked on October 31, 2026. Letter counts and recommender rules change between cycles, so recheck the program page before you upload. Columbia's Data Science Institute, NYU's Center for Data Science and the University of Washington's MSDS pages blocked or did not answer our automated fetch that day, so they are not quoted.

On this page: what programs ask · the caught-error paragraph · evidence by contact · constructed sample · manager letters · checklist · questions

What data science programs ask recommenders

Three data science master's programs publish recommender guidance that goes beyond a letter count. Read the last column as the brief your letter answers.

Three details matter to a recommender:

• Rice lets a strong letter offset a weak record. Its page says "Admissions exceptions can be made if a student has sufficient work experience, GRE scores that show a high level of competence, a master's degree, strong letters of recommendation, and/or specific circumstances that led to a low GPA." For applicants with a GPA under 2.9, it lists GRE scores, several years of relevant work experience and strong letters as things that "are highly recommended" (another line on the same page says the GRE is required below a 2.9 GPA "unless an exception is granted by the Rice Admissions committee"). If your student's marks are weak, your letter is doing more work than usual.
• Rice decides on a rolling basis, and your letter can be the bottleneck. Applicants "can generally expect an admission decision within approximately four to six weeks after all required application materials, including letters of recommendation, have been received." The Spring 2027 application deadline on the page is November 1, 2026.
• UChicago does not need your letter at submission. "Letters of recommendation can be uploaded after you submit your application." Its first in-person priority deadline is November 9, 2026, and the international deadline for applicants needing UChicago visa sponsorship is January 26, 2027.

What none of the three pages publish: how much a letter weighs against grades, or how a reader rates it. Anything you read about letter "weightings" for these programs is a guess.

The data mistake the student caught is your best paragraph

Our view: the single most useful paragraph in a data science letter describes a moment when the student found something wrong with data or an analysis, and what they did about it. Leakage, a biased sample, a metric that rewarded the wrong thing, a join that duplicated rows. Then say what changed because they caught it.

The basis: Rice asks for people familiar with the applicant's "interest in data science and analytics", not their programming. Duke says applicants with strong quantitative experience "should demonstrate their ability to communicate, solve problems, and work effectively in teams". A caught error shows quantitative judgment, problem solving and communication in one incident. A list of tools shows none of them, and the CV already has it.

The countercase: if you only taught the student in a large lecture course, you probably never saw this kind of moment. Do not invent one. Write what you did see (a project grade, a viva answer, a question they asked) and say plainly that your view is from the classroom. A letter that overreaches is worse than a short one. Also, if the student is applying to a research-track program in machine learning rather than an applied data science master's, the LOR sample for MS in Computer Science is closer to what those readers want.

Keep three things apart when you write. Published: the asks in the table above. Our recommendation: the caught-error paragraph. Unknown: how any committee weighs it.

Which evidence to use, depending on what you saw

Pick the row that matches your contact with the student. Write one incident from that row, with numbers you can stand behind.

If you can rank the student, name the group: the strongest of the 46 students in my 2025 regression course is a comparison a reader can use. One of the best students I have taught is not.

Sample LOR for MS in Data Science (constructed)

Constructed example: statistics professor, weak and revised

Constructed example, written by us, not a real recommendation letter. Context (invented): Professor R. Iyer taught Ananya Rao two statistics courses and supervised her final-year project at an Indian university. Ananya is applying to US data science master's programs.

The weak version (65 words):

Ananya Rao was a student in my Probability and Regression courses. She is very hardworking, sincere and intelligent. She is proficient in Python, R, SQL and machine learning libraries. Her final-year project on customer churn used advanced analytics to find the drivers of churn. She has a strong interest in data science and will be an asset to your program. I recommend her without reservation.

It names tools the CV lists, a project with no finding, and three adjectives with no incident.

The revised letter (324 words):

I am Professor of Statistics at [University], and I recommend Ananya Rao for your MS in Data Science. I taught her in two courses, Probability (2024) and Regression Analysis (2025), and supervised her final-year project from July 2025 to March 2026. I have known her for two years.

In Regression Analysis she placed first of 46 students on the final examination. The marks matter less to me than a question she asked in week six, when she pointed out that our textbook example assumed independent observations in data collected from the same households over time. She was right, and I changed the example the following term.

Her final-year project used a local telecom partner's customer survey to find which service problems made customers leave. Her first analysis said billing complaints mattered most. She came to me worried rather than pleased. The survey had been sent only to customers with active accounts, so the customers who had already left were not in the data at all. She asked the partner for its account-closure records, matched them to the complaint log, and showed that among customers who had actually left, network outages came first and billing came third. She then wrote a one-page note for the partner's analysts, in plain language and with one chart, explaining why the survey alone pointed the wrong way. Their team lead told me it was the first student report they had actually used.

She is not yet strong in software engineering. Her code works but is not well tested, and I have told her so. A structured master's program is the right place to fix that, and she knows it.

In twenty-two years of teaching I have supervised about 60 final-year projects. Ananya's is one of the three I would show a new colleague as an example of careful work. I recommend her strongly and without qualification for graduate study in data science.

Professor R. Iyer, Department of Statistics, [University]

Why the revision works, against what the programs ask:

1. It establishes the vantage. Two courses, one project, dates and duration. A reader knows exactly what Professor Iyer could see.
2. The survey paragraph is the evidence Rice and Duke ask for. It shows "interest in data science and analytics" as behaviour, and it shows problem solving and communication (the note for the analysts) in one incident.
3. Who used the result is the point. The weak version names a project. The revised version shows a sample that left out the people who mattered, the fix, and a non-technical team acting on the corrected answer. Duke asks applicants to show they can "communicate, solve problems, and work effectively in teams"; this paragraph covers the first two, and a reader cannot get it from the CV.
4. Every comparison names its group: 46 students in one course, about 60 final-year projects over a career.
5. One honest limitation makes the praise credible. It is specific, and it says what the degree will fix.

The numbers and the telecom partner in the letter are invented. Use your own.

Professional letters for career changers

UChicago's page strongly recommends at least one letter from "a direct manager/supervisor or internship supervisor who can attest to skills you demonstrated or gained though [sic] a professional workplace experience (e.g., leadership, teamwork, collaboration, initiative, management, other)." Rice lists "direct supervisors or people in similar roles" first. For an applicant who has worked as an analyst, an engineer or in a non-technical role, the manager's letter is not a fallback. At UChicago it is the recommended shape.

For the statement that goes with a manager's letter, see the engineer to data science SOP.

A manager's version of the caught-error paragraph names the business question, the data the applicant used, the mistake they found or owned, and the decision that changed. Name your team size and how many people you have managed, so best analyst on my team has a group. The employer letters of recommendation guide covers who at a company should write and the email-domain questions.

If the applicant is moving into data science from a non-technical background, Duke's line applies: they "should show their readiness to build those skills, whether through coursework, professional experience, or recommendations." A manager can write that readiness paragraph: the spreadsheet model the applicant built without being asked, the online statistics course they finished while working full time, the report they automated. The applicant's side of that case is in the data science master's SOP guide for non-technical backgrounds.

Before you upload a data science letter

1. One sentence states how you know the student, in what role, and for how long.
2. At least one paragraph describes a data problem the student handled, in their own actions.
3. No paragraph lists tools or libraries unless the paragraph is about what they did with one.
4. Every number in the letter is one you can stand behind if the program contacts you.
5. Every "best" or "top" names its group and its period.
6. The endorsement is stated plainly in the last paragraph.
7. You checked the program's page for the letter count and whether it wants a workplace supervisor.

Questions about data science recommendation letters

Writing for an engineering department instead? The LOR samples for MS mechanical and electrical engineering cover the failed-test paragraph and TA evidence.

Should a data science LOR come from a professor or a manager?

It depends on the program. UChicago strongly recommends at least one from a manager or internship supervisor. Rice accepts supervisors, professors and research advisors. Our view: a recent graduate should ask the professor who supervised their project; someone with work experience should send one of each, so both kinds of evidence are covered.

How long should the letter be?

None of the three programs states a word limit. The constructed letter above is 324 words. The graduate school letters of recommendation guide covers length norms and what programs publish about them.

Can the student draft the letter for me to sign?

None of the three program pages addresses this. GradPilot's letter reviews aren't for a letter you drafted for a recommender to submit as their own. If you are the recommender, write it yourself and use the student's CV and SOP for facts and dates.

Is this different from a computer science LOR?

Yes, in emphasis. A CS letter for a research-track program leans on design and debugging. A data science letter leans on judgment about data and on communication with people who use the results. The LOR sample for MS in Computer Science covers that letter, and the LOR sample for MS in Business Analytics covers programs that want manager answers to set questions.

Get a free review of the letter

Get a free Quick Review of your recommendation letter against GradPilot's graduate school recommendation letter rubric. It reads how you know the candidate, incidents instead of adjectives, comparisons that name their group, the case for graduate potential, and the strength of your endorsement. Paste the letter and name the program you are writing for.

The applicant's own document is covered in the data science SOP guide, and the graduate school essays hub lists the rest.

Sources

Retrieved October 31, 2026, except where noted.

• Rice University, Online Master of Data Science Admission Requirements (Spring 2027 deadline listed): two letters, who should write, the GPA exception and the rolling decision timeline.
• University of Chicago Data Science Institute, MS in Applied Data Science: How to Apply (Autumn 2027 entry): two letters, the workplace-supervisor recommendation, upload after submission, deadlines.
• Duke University, Admissions - Duke MIDS (no cycle stated): what applicants with strong and with less quantitative experience should show.