🎯 Final Project Guidelines

STAT 266A, Fall 2026

Author

The final project is 60% of your grade. You analyze a data-driven research question of your choice, in R, and build the analysis up across the quarter through five deliverables. This page includes everything you need: what each deliverable is, how it is graded, how pairs work, and exactly how AI may (and may not) be used.

The idea behind the design

Anyone can now get code that runs. What this course assesses is whether you understand the analysis: why this plot, why this model, what could be wrong with it. So every deliverable includes a conversation about your work, and each partner in a pair is assessed individually. Details below are draft defaults for Fall 2026; any change will be announced on Ed Discussion.

📅 Timeline

Week	Date	Deliverable	Weight	Format
4	Tue, Oct 20	Proposal	10%	Oral presentation
6	Tue, Nov 3	Mid-quarter report	10%	Written (Quarto)
7	Thu, Nov 5 - Tue, Nov 10	Mid-quarter check-in	5%	5-minute individual conversation
9	Tue, Nov 24	Peer review	5%	Written review (no AI)
10	Tue, Dec 1 and Thu, Dec 3	Final presentation	15%	Oral exam: presentation + individual Q&A
Finals week	Dec 7-11	Final written report	15%	Written (Quarto) + AI reflection

Each deliverable is hard-deadline, with one chance to improve after it is graded.

👥 Working in a pair

Pairs are held to the same standard as individuals, so:

Both partners must understand everything. In every oral component (proposal, check-in, final Q&A), each partner is questioned separately and may be asked about any part of the project, including code the other partner wrote.
Shared GitHub repository. Both partners commit to it throughout the quarter. Commit history is how I see the project evolve, and it is reviewed when contribution is in question.
Contribution statement. With each written report, submit a short paragraph per partner: who did what, and what each of you learned. Be candid; an uneven split that is honestly reported is fine.
Grades can differ within a pair. The individual components (Q&A, check-in, contribution) are graded per person.

If a pair is not working, talk to me early, before a deadline.

🧰 Technical requirements (all deliverables)

The project is also a reproducibility exercise. From the first deliverable on:

Work in a Positron project with a clear folder structure (data, code, output).
Track it with git and push to a GitHub repository (a free account is enough). Commit early and often, with messages that say what changed.
Report in Quarto (.qmd) so that anyone can re-render your analysis from a clean session: no hidden state, no absolute file paths.
Document your data: where it came from, its license or terms of use, and any cleaning you did.
Follow the readable-code practices from Week 9: meaningful names, comments that explain why, short functions.

🤖 AI on the project

The project is the one place where AI is allowed, with disclosure (see the course AI Use Policy). The rule of thumb is that AI may help you work, but it may not do your understanding for you.

Generally fine (disclose it)	Not okay
Having an error message explained	Generating your whole analysis, then submitting it without understanding it
Asking for approaches, then choosing among them yourself	Using AI in the oral components: proposal, check-in, Q&A
Style or readability suggestions on code you wrote	Using AI to write your peer review
Spell-checking or tightening your own prose	Pasting AI-written interpretation of results you have not verified
Debugging help on code you wrote	Copying package names, citations, or “facts” you have not checked

Verify everything. AI-written code frequently runs but is wrong. Before it enters your project: check row counts before and after joins and filters, look at the data and the plot itself, check regression assumptions and residuals, and confirm every function and package actually exists and does what the AI claims. Check any citation it gives you. You are responsible for every error, bias, or unsupported claim in what you submit.

The test. If I ask you about any line of your code or any sentence of your interpretation, you should be able to explain it in your own words without a screen in front of you.

AI disclosure template

Put this at the end of every deliverable (a short section titled “AI use” in a Quarto report; on the last slide for a presentation):

## AI use

- **Tools used:** (e.g., "Claude", "Copilot in Positron", or "None")
- **What I used them for:** (e.g., "explained a `left_join` warning", "suggested a
  faceted layout for Figure 2")
- **What I changed or rejected, and how I checked it:** (e.g., "the suggested model
  ignored the repeated measures; I fit a mixed model instead and compared residuals")

“None” is a complete and welcome answer. Honest disclosure is never penalized; undisclosed use is an academic integrity issue.

📝 The deliverables

Proposal

Oral presentation, Week 4 (Tue, Oct 20). 10%. About 3 minutes plus questions.

Cover:

Question: the data-driven question you want to answer, in one or two sentences.
Data: what dataset(s) you will use, where they come from, and their size and variables. Show the first rows.
Plan: which visualizations and which model(s) you expect to use, and why.
Risks: what could go wrong (data quality, too few observations, a question the data cannot answer).
Team: working alone or in a pair.

Upload your slides to Canvas as one PDF by 1:00 pm on Tue, Oct 20, so all presentations run from one laptop.

Each partner answers questions. No AI while presenting or answering. Include a short AI-use note on your last slide.

Look for: a question the data can actually answer; a dataset you have already opened; a plan that is realistic for the quarter.

Mid-quarter report

Written report, due Tue, Nov 3 (10%), plus individual check-in Nov 5-10 (5%).

Submit a rendered Quarto report and your GitHub link, containing:

The refined question and a description of the data.
Data cleaning steps, with a note on what you checked (row counts, missing values, oddities).
At least two well-designed ggplot2 figures with informative titles, labels, and captions, with a sentence on what each figure shows and why you chose that type of plot.
A first regression model with a clear interpretation of the coefficients and a check of its assumptions.
A contribution statement (pairs) and an AI use section.

Check-in (mandatory). Each person has a 5-minute slot during Week 7 (Thu, Nov 5 or during office hours). There is no sign-up: the check-in schedule is posted on Ed Discussion. Each person, in a pair separately, answers questions such as: “Walk me through this pipeline.” “Why this model rather than another?” “What happens to this plot if that observation is removed?” No screen sharing of AI; your own code and report only.

Look for: clarity, correctness, and that you can defend every step yourself.

Peer review

Written review, due Tue, Nov 24 (Week 9). 5%. Written by you, no AI.

You are assigned a classmate’s project (repository and presentation). Review:

Code: Does it re-run from a clean session? Is it readable? Are there hidden assumptions or errors? Use the Week 9 code-review checklist.
Presentation: Did the question and finding come across? Were the figures clear?

Write specific, kind, actionable feedback: two strengths, two concrete suggestions, and one question you would want answered. Comments like “looks good” are not enough. Your review is graded on how useful it is, not on whether you found the most flaws.

Final presentation

Oral exam, Tue, Dec 1 (work session) and Thu, Dec 3 (in class). 15%. About 5 minutes to present plus individual Q&A.

Present your question, main findings (with your best figures), the model, and its limitations.
Then each person answers questions separately about their project’s code, analysis, and earlier deliverables: the parts you wrote and the parts your partner did.
No AI. You may bring your rendered report or code as reference.
Slots are posted on Canvas and Ed Discussion in Week 9.

Look for: you understand your own analysis, can say what it does and does not show, and can explain your decisions.

Final written report

Due during finals week, Dec 7-11, 2026. 15%.

A complete Quarto report, with a clean re-render and a GitHub repository, that:

Incorporates the feedback from your proposal, mid-quarter report, peer review, and presentation, and says how.
Presents the full analysis: question, data, cleaning, visualizations, model(s), interpretation, and limitations.
Ends with three required items:
1. AI use (template) for the whole project.
2. AI reflection (about a paragraph or two): where did AI help you, where did it mislead you or cost you time, and what would you do differently? Reflect honestly, including on whether it helped you learn.
3. Contribution statement (pairs).

Repository check: the analysis must re-run from a fresh clone, with a README explaining how.

🔁 Revisions

After each deliverable is graded, you get one chance to improve it. Resubmit the revised work with a short reflection: what you changed, why, and what you learned. For oral deliverables the revision is a short written follow-up answering the feedback questions; for the final presentation it is submitted with the final report. This exists so that a missing piece becomes a learning opportunity instead of a permanent mark.

⚖️ How it is graded

Each deliverable has a rubric posted with its assignment. Across all of them I look at:

Understanding: you can explain and defend your choices.
Correctness: the code, analysis, and interpretation are sound, and verified.
Communication: figures and writing are clear and honest about limitations.
Reproducibility: it re-runs; the repository is organized and documented.
Honesty: AI use and contributions are accurately reported.

🙋 Getting help

Come to office hours (after class Tue and Thu), post on Ed Discussion, and use the Tuesday work session in Week 10. Ask about anything unclear before the deadline. This includes whether something is allowed under the AI policy.