How much of school outcomes is just demographic composition?
A reproducibility log: the data this analysis touched, the queries it ran, the outside sources it leaned on, and the steps a third party would follow to re-run it.
Summary
Regressed school-level academic composite on proclivity_score (continuous version of decile) to show how much variance in outcomes is explained by student-body composition alone.
Data sources
- Database tableschools
One row per New York City public school (DBN, name, district, borough, school_type, grade_band, latest_enrollment, proclivity_decile, proclivity_score, closed_at, admission_category).
- Database tableschool_year_metrics
Long-format per-school per-year metric facts (school_dbn, year, metric_key, subgroup, value, suppressed). Loaded from DOE/NYSED public files via scripts/loaders/*.
- Database tableproclivity_score / proclivity_decile
Composite student-body-challenge score (lower = lower-need, higher = higher-need) built from % economically disadvantaged, % English Language Learners, % students with disabilities. Decile 1 = lowest-need; decile 10 = highest-need.
Steps
Per-school 2024-25 composite vs proclivity_score; compute Pearson R² and a scatter.
Caveats
R² depends on the outcome metric and the school subset (ES, MS, HS). Different choices change the headline number.
Reproduce
Clone the repo, set DATABASE_URL to a Postgres with the project schema loaded, run `npx tsx scripts/loaders/<source>.ts` for any not-yet-loaded data, then issue the queries in the Steps section. The story page also lists the exact `metric_key`/`subgroup`/`year` filters used. Searchable by the answer's headline number — every figure is recomputable from the queries shown.
The recipe lives at data/stories/recipes.ts in the repo. Corrections welcome.