Data guidelines

How to choose appropriate datasets for course work

Short, practical guidance on where the data in your course assignments should come from.

The short version

Use data from one of these four categories:

1. Built-in R datasets (mtcars, iris, palmerpenguins, gapminder, ggplot2::diamonds, dplyr::starwars, etc.).
2. Simulated data generated in-script with set.seed().
3. Public datasets with verifiable provenance and license — OpenIntro, TidyTuesday, official statistical agencies, journal supplementary materials with clear reuse terms.
4. Instructor-approved student-selected datasets — bring your own, but check with me first.

If your data doesn’t fit one of these four buckets, find different data. There’s no shortage of clean public data on the topics this course cares about.

What “good dataset” means here

A good dataset for a course assignment is:

• Tidy enough that you can spend your time on the analysis, not on shape-fixing.
• Documented — there’s a real codebook, README, or paper that defines each column.
• Appropriately sized — small enough to load fast, large enough to support a real plot.
• Cleanly licensed — you can use, share, and submit work derived from it without ambiguity.

What we do not use

• Old archive datasets from prior terms of this course. These are off-limits. The course’s data sources are listed above; the R Project brief in the course LMS echoes the same four categories.
• Datasets you can’t cite the source of. If you can’t find the origin, the license, or the codebook, don’t use it.
• Datasets containing identifiable people without explicit permission and clear public-use terms.

Recommended starting points

• Built-in R. Run library(help = "datasets") in R, or data(package = "palmerpenguins") after loading a package.
• OpenIntro. https://www.openintro.org/data/ — categorized, documented, and clearly licensed under Creative Commons.
• TidyTuesday. https://github.com/rfordatascience/tidytuesday — weekly curated datasets aimed at R learners. Each week has a README, a codebook, and source attribution.
• Government / statistical agencies. US Census, BLS, CDC, WHO, OECD, etc. — usually open, usually documented, sometimes large.

If you bring your own

Send me three things before you commit to the dataset:

1. The source link (the canonical page, not a re-shared mirror).
2. The license or terms of use.
3. A one-sentence description of what the rows and columns are.

I will say yes to most things in the first sentence. The check is about catching ambiguous licensing or accidentally-identifiable data before you build a project around it.

Privacy reminder

If your dataset contains anything that could be tied to an individual — names, exact addresses, fine-grained health information, identifiable workplace data — do not paste it into an AI assistant, and do not put it in your portfolio. See the AI use guidelines for what to share and what not to share.