Quantitative Analysis of Local Singles Market
A reproducible data report on fictional Null Island: frame the population, audit the denominators and explain a misleading bio comparison.
Due: 19 March 2027 · Weight: 20%
The brief · data report#
A fictional product manager wants to launch a dating service on Null Island. Their memo reads: “Half the local singles are available, one in four is a match, and bio A is the winner. Scale immediately.”
Use the market census and bio-exposure files to audit those three claims. They describe different frames and measurements. Your job is to say which statements the supplied data supports, which need qualification, and which cannot be inferred. This is a report about an invented dataset, not a survey of people in your city.
Required analysis#
For rehearsal, use The Experiment That Lied to inspect the exposure reversal and record its limitations. Its downloadable note can support your Bio experiment protocol; your submission still needs the complete analysis and reproducible working below.
- Define active, available and reciprocally eligible populations. Reproduce 150/300, 40/150 and 40/300, with units and denominators.
- Audit row constraints and show the effect of adding 50 missing active profiles to East while leaving reciprocal counts unchanged.
- Compare both bios in aggregate and within each zone. Explain why the pooled winner is not a demonstrated causal winner.
- Propose a stratified randomised follow-up experiment with one primary outcome, a fixed stopping rule and a minimum worthwhile effect.
- Distinguish sampling uncertainty, model uncertainty and the absence of real-world external validity. Do not attach a population confidence interval to an invented census and call it research.
Use the Platform audit, Photo asset manifest and Bio experiment protocol as your preparation. The photo manifest helps explain why an experiment must hold non-bio inputs constant; a candidate profile is not a report requirement.
What you submit#
One 1,000–1,400-word PDF or accessible HTML report; at most three clearly labelled figures; a notebook or script plus its data and a README giving one reproducible run command. A spreadsheet is accepted if formulas and an exported calculation log are included. Identify the supplied data as synthetic on every figure.
Append a short tool-use and source note outside the word count. Quote counts directly from your calculation, not from generated prose. Do not scrape dating apps, estimate classmates’ eligibility, or include real profiles or messages.
Submission and feedback#
Due 19 March 2027, 5 pm Canberra time. In this fictional course, bring the report and source bundle to Casimir on removable media or arrange an accessible digital handover at consultation. This website does not upload or receive submissions. Written feedback returns within seven days.
The published criteria total 100 points within this assessment; this report contributes 20% of the course. Keep the feedback: the final model card must explain how it changed your handling of data or claims. Policies cover extensions and review.
The spec
Use this checklist before handing in your work. The marking criteria below explain how your decisions and communication will be assessed.
- Use only fictional cases and synthetic data; no real romantic outcome is graded.
- Include reproducible working and source attribution under the assessment's stated tool rules.
How it is marked
| Criterion | Weight |
|---|---|
| Population and denominator definitions | 30% |
| Reproducible calculations and data validation | 30% |
| Experiment design and limitations | 30% |
| Clear figures and source attribution | 10% |