The DOI matches its Crossref record; 5 of the 5 references it cites were found in a registry or an index; 2 of the 8 sources its claims cite were checked for support.
Generative AI for Data Science 101: Coding Without Learning to Code
Open accessAbstract
Should one teach coding in a required introductory statistics and data science class for non-major students? Many professors advise against it, considering it a distraction from the important and challenging statistical topics that need to be covered. By contrast, other professors argue that the ability to interact flexibly with data will inspire students with a lasting love of the subject and a continued commitment to the material beyond the introductory course. With the release of large language models that write code, we saw an opportunity for a middle ground, which we tried in Fall 2023 in a required introductory data science course in our school’s full-time MBA program. We taught students how to write English prompts to the artificial intelligence tool GitHub Copilot that could be turned into R code and executed. In this short article, we report on our experience using this new approach.
Overview
CoChat Summary
TL;DR
The authors describe replacing point-and-click software with GitHub Copilot in a required introductory data science course for full-time MBA students, teaching English prompting instead of R syntax. They report that this "translator-in-your-ear" approach let non-coders interact flexibly with data while preserving statistical learning objectives.
Key findings
- The authors rate six approaches to software in data science courses across five characteristics (work with data, easy to teach, extends to methods not covered, generalizes to other languages, used in practice); Copilot is the only approach favorable on all five (Table 1).
- Prompting is itself a teachable skill, distilled into three principles: be specific, think about the context window, and break complex operations into simpler steps.
- Copilot's outputs are random: prompts that worked in class preparation sometimes failed live, and with 60+ students following along, "one could often have 20 students getting something different from the rest."
- A student who had never programmed before, and entered the semester uninterested in the course, discovered on his own that rounding predictions to the nearest integer substantially reduced test error in the wine-quality regression exercise — a function ("round") never mentioned in class.
Methods
- Design: a teaching-practice/experience report on a Fall 2023 iteration of "GSBA-524: Data Science for Business," a compulsory first-semester full-time MBA course at USC's Marshall School of Business, with three sections of roughly 65 students each, 31 sessions of 1.25 hours, six homework assignments, six quizzes, a midterm, and a final exam.
- Intervention: the course topics and structure were left unchanged, but the prior Radiant point-and-click platform was replaced with GitHub Copilot used as an English-to-code translator; students installed R, Visual Studio Code, and the R and Copilot extensions, later moving to RStudio's Copilot integration.
- Data/analysis: qualitative reporting of challenges and benefits observed across the semester, plus a worked Copilot session walkthrough (Appendix B) illustrating Principles 1 and 2; no quantitative outcome measures are reported.
Limitations
- The authors state they do not necessarily recommend the approach in other contexts: in an elective for students with the background, time, and interest to learn to code, Copilot "could undermine a student's learning process" (analogous to giving calculators before arithmetic is learned).
- The authors acknowledge the objection that students who cannot read code cannot verify its correctness, and concede their remedy — prompting for checks and scrutinizing plots — is "an imperfect process that is unlikely to be sufficient for complex programming projects."
- The authors note Copilot's lack of transparency about its inner workings (no definitive documentation of the context window) and that it does not allow offline coding; evident from the text is that the report rests on a single semester, a single institution, and one instructor team's impressions rather than controlled comparison.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
2 Valid · 4 Not valid · 6 Not checked · 54 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“One notable form of inconsistency in code generated by Copilot is whether it opts for tidyverse (Wickham et al. 2019) or base R in a given analysis.”
§ plot price versus number of bathrooms for all the condos. > 3.1 Challenges · p. 7
What the sources say
- Supportedcites Wickham et al. 2019 source
The citation is attached to “tidyverse,” identifying the package collection that the citing authors contrast with base R. The source describes the tidyverse and its component packages (Components) and explicitly distinguishes its priorities from base R (Abstract). The claim about Copilot’s inconsistency is the citing authors’ observation, not a finding attributed to this source. (LLM, conf=0.94)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“that Copilot did not have access to the data set itself, even when loaded in the environment”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“we found prompts that 'worked' earlier in the day would sometimes start generating different and sometimes worse responses.”
§ plot price versus number of bathrooms for all the condos. > 3.1 Challenges · p. 6
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“One challenge that was particularly disconcerting to students and us alike was grappling with the fact that Copilot's outputs are random.”
§ plot price versus number of bathrooms for all the condos. > 3.1 Challenges · p. 6
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“When we run this code, we get the desired plot:”
§ plot price versus number of bathrooms for all the condos. > B A Session with Copilot Demonstrating Principles 1 and 2 · p. 16
plainstates a causeSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
10 more claims
Create a free account to see all 66 claims and every source verdict.
- Probably not a substantial claim. 43
- The quote only partly states this claim. 1
- It may state two separate claims as one. 14
Citations
Cites
Wickham, H., Averick, M., Bryan, J., Chang, W., McGowan, L., François, R., Grolemund, G., Hayes, A., Henry, L., Hester, J., et al. (2019) · Journal of Open Source Software · doi:10.21105/joss.01686 · Cited by 22,138
Modeling wine preferences by data mining from physicochemical properties
Cortez, P., Cerdeira, A., Almeida, F., Matos, T., & Reis, J. (2009) · Decision Support Systems · doi:10.1016/j.dss.2009.05.016 · Cited by 1,273 · cited 4×
“For example, we used the well-known “wine” dataset from the UCI data bank (Cortez et al. 2009b).”
InfluentialEllis, A. R., & Slade, E. (2023) · Journal of Statistics and Data Science Education · doi:10.1080/26939169.2023.2223609 · Cited by 137 · cited 1×
“Most of this work has centered on the use of ChatGPT in the classroom (see, e.g., Ellis and Slade 2023, and references therein).”
Paulo Cortez, A. C. (2009) · UC Irvine · doi:10.24432/c56s3t · Cited by 42
A Tutorial on Teaching Data Analytics with Generative AI
Bray, R. (2023) · doi:10.2139/ssrn.4484395 · Cited by 4 · cited 1×
“One of the most relevant examples to our own is Robert Bray’s Spring 2023 class at Northwestern, also for MBA students, in which he had the students use ChatGPT for generating R code (Bray 2023).”
Cited by
Students’ Statistical Thinking When Using Generative AI: A Descriptive Case Study
Rao, V. N. V., Jeng, A., Drew, J., Kala, B., & Gongati, S. (2025) · Journal of Statistics and Data Science Education · doi:10.1080/26939169.2025.2596684 · Cited by 9
“Pushing beyond Bien and Mukherjee (2025), by investigating students’ thinking, we uncovered new insights into how students’ statistical content knowledge interacts with their AI procedural knowledge in the completion of statistical tasks.”
InfluentialDu, Y., Zhao, H., Zhu, X., Zhang, H., & Liu, X. (2026) · International Journal of Human-Computer Interaction · doi:10.1080/10447318.2026.2622584 · Cited by 5
“Furthermore, moving beyond the predominant focus on cognitive tasks like writing or coding (e.g., Bien & Mukherjee, 2025; Ghafouri et al., 2024), this study extends the discussion to oral communication practice, a scenario that more closely mirrors the dialogic affordances of GenAI.”
A Tutorial on Teaching Data Analytics with Generative AI
Bray, R. (2023) · doi:10.2139/ssrn.4484395 · Cited by 4
ÖZGÜL, E. (2026) · Kastamonu Eğitim Dergisi · doi:10.24106/kefdergi.1878122 · Cited by 6
Dalzell, N. M., Rehnberg, Z. L., & Theobold, A. S. (2025) · Journal of Statistics and Data Science Education · doi:10.1080/26939169.2024.2449159
Metadata
- Authors
- Bien, J., & Mukherjee, G.
- Affiliations
- University of Southern California (US)
- Corresponding author
- Bien, Jacob
- Venue
- Journal of Statistics and Data Science Education
- Publisher
- Informa UK Limited
- Published
- 2025-01-27
- Volume
- 33
- Issue
- 2
- Pages
- 129-142
- ISSN
- 2693-9169 (ISSN-L)
- OpenAlex
- W4404624965
- Cited by
- 15
- References
- 5
- Citations by year
- 2026: 10 · 2025: 4
- FWCI
- 4.84
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Licence
- Locations
- Journal of Statistics and Data Science Education (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Mathematics › Statistics and Probability › Statistics Education and Methodologies
- Physical Sciences › Computer Science › Computer Science Applications › Teaching and Learning Programming
- Physical Sciences › Mathematics › Statistics and Probability › Data Analysis with R
- Keywords
- generative AI, large language models, GitHub Copilot, non-major students, prompt engineering, R programming language, MBA students
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-10-28, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-28 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Generative AI for Data Science 101: Coding Without Learning to Code. https://app.cochat.ai/papers/10.1080/26939169.2024.2432397
- The paper
- Bien, J., & Mukherjee, G. (2025). Generative AI for Data Science 101: Coding Without Learning to Code. Journal of Statistics and Data Science Education. https://doi.org/10.1080/26939169.2024.2432397
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.