Week 3 · Suggested 60 minutes · First-pass draft
A survey with a moving denominator
Learn why large response counts cannot rescue a population estimate when groups respond at very different rates.
Applied assignment, not a stored grade. Instructions and model checks are provided for self-study or instructor adaptation. Work is not transmitted to this site.
What you will do
- Use the selection-bias laboratory with fixed 60/40 population groups and support of 70% versus 30%.
- Predict the expected respondent support when Group A responds at 90% and Group B at 30%.
- Alter Group B's response probability to 90%, then to zero, while the population opinions remain fixed.
- Write a recommendation for what a survey organization should report about the respondent mix.
What to submit
A small comparison table and a short methodological critique of why respondent-weighted rates may be misleading.
Self-check before finishing
The underlying town support stays at 54%; its unweighted respondent rate need not.
Assigned reading
Distinguish descriptive association from population inference; inspect selection and clustering.
- Lesson 9: association and correlation
- Lesson 10: sampling and uncertainty
- Lesson 11: stratified cluster and complex samples
Instructor discussion guidance
Emphasize that counts in the demonstration are expected values rather than observed or simulated survey results.
Print a predict–experiment–explain worksheet → · Open browser R →