Opening
The ultimate statistical accomplishment is not a tiny p-value. It is a report that another person can reproduce, interrogate, and perhaps disagree with intelligently.
Teaching narrative
Start with a question whose scope matches your evidence. Define the observation, data source, measurement units, transformations, missingness, and research design before choosing a procedure.
Explore the data visually. Specify a justified model and report its estimates, uncertainties, and assumptions. If the observations are not randomized experimental treatment assignments, explain why you are reporting adjusted associations instead of automatically claiming causal effects.
Run at least one sensitivity analysis. What changes when an influential observation is retained, a predictor is added, the time period is restricted, or the model form changes? Record these choices, including analyses that did not produce exciting results.
Deliver the R code, figures, source documentation, limitations, and a clear report. If a reader cannot reproduce the output from the supplied instructions, the exercise is not finished. The capstone asks you to show that you can reason scientifically, not just remember where the buttons are.
Before the experiment
What is the strongest statistical claim an observational air-quality regression can support without additional causal identification assumptions?
Live demonstration
Use the R airquality field notebook: count missing data, fit two models on the same complete-case rows, and inspect residuals.
Exit question
Write the single limitation you would most want a skeptical reader to notice before interpreting your conclusion.