New case study: Epidemiology risk ratio with strata (skeleton)

[nicholaskarlson]

Summary

No response

Details

Summary

Add a new Chapter 15/16 style case study for epidemiology:

• Simulate a simple binary exposure / binary outcome dataset with strata
  (e.g., age group).
• Provide an analysis script that computes:
  • crude risk ratio,
  • stratum-specific risk ratios,
  • and a pooled / adjusted risk ratio (e.g., Mantel–Haenszel).

This issue is for building the basic simulator + analyzer skeleton.
We can refine the statistics and narrative in later issues.

Files to add

• scripts/sim_epi_rr_strata.py – simulator for stratified 2×2 tables
• scripts/chXX_epi_rr_strata.py – analyzer (chapter number TBD; use XX for now)
• Makefile targets:
  • epi-rr-ci (small n, CI smoke)
  • epi-rr (full demo)
• Tests:
  • Add the new scripts to tests/test_cli_smoke.py.

Goals

• Simulator: generate a small dataset with columns like:
  • id, stratum, exposed (0/1), outcome (0/1)
• Analyzer: read a --data or --datadir input and compute:
  • crude risk ratio with a confidence interval (basic is fine),
  • per-stratum 2×2 counts and risk ratios,
  • a placeholder for pooled/adjusted RR (even if initially just a TODO + rough calculation).
• Use the shared CLI helper (scripts/_cli.py) for argument parsing.
• Wire the new scripts into the Makefile and smoke test.

Hints

• Look at the existing pattern for Ch13/Ch14/Ch15:
  • simulator script → CSV in data/synthetic/
  • analyzer script → summaries/plots in outputs/...
• For a first pass, you can use simple formulas from standard epi texts,
  or even rely on statsmodels if convenient.
• Keep everything small, clear, and easily extended in follow-up issues.

Difficulty

Medium: good first issue for someone interested in biostats/epidemiology
and comfortable with basic Python and 2×2 tables.

Files to Touch

No response

Contributor Checklist

• I have read CONTRIBUTING.md.
• I can run make lint locally.
• I can run make test locally.
• I have checked for existing issues/PRs that might overlap.