Labs

STAT 45203 · short resampling & robustness activities

Public, ungraded lab exemplars — instructor-authored, synthetic data. The graded lab prompts, point values, and due dates live in Blackboard. Open texts are cited by section only; no code or prose is copied. See Open readings & attribution.

These labs connect a method’s logic to computation. Each is short and self-contained: a data setting, a small amount of R you can run in Posit Cloud, a picture to interpret, and a written conclusion that names what the method assumes and what it protects against. They mirror the graded labs in shape but are practice — run them, break them, and check the output against the weekly notes.

Every lab expects the same three habits from the AI policy: say what is being resampled or ranked, what stays fixed, and how you verified the output.

The labs

  1. Permutation test from scratch — build a null distribution by shuffling group labels and read a p-value off it. Pairs with Weeks 3–4.
  2. Bootstrap confidence intervals — resample with replacement, form a percentile interval, and see where it disagrees with a t-interval. Pairs with Weeks 5–6.
  3. Rank-based methods — signed-rank and rank-sum tests, and what ties and ranks throw away. Pairs with Weeks 7–8.
  4. Robust summaries & outlier sensitivity — mean vs. median vs. trimmed mean under a single contaminating point. Pairs with Week 10.
  5. A small simulation study — measure Type I error and power for two competing tests under conditions you control. Pairs with Week 13.

How to run them

You can run every lab in a browser with Posit Cloud — no install required — or locally in RStudio. See Software setup. Figures on these pages are static (generated ahead of time); the R shown is what you would run to reproduce them.