Perception gapsFull briefing
How much of school outcomes is just demographic composition?
The longer form of the briefing for this story: the original thesis, the supporting findings the data team flagged, and the reporting directions a journalist could follow. For the short version and the data analysis, see the main story page.
Thesis
Our proclivity decile (built from % econ-dis, % ELL, % SWD) explains a huge chunk of NYC schools' average academic outcomes — about 50% of the variance in grade-5 math proficiency. The other 50% is the school. Distinguish those two halves and you have a much fairer evaluation framework.
Supporting findings
- Proclivity-decile groups and avg ES composite (2024-25): decile 1 = 86%, decile 5 = 63%, decile 10 = 42%.
- R² of proficiency ~ proclivity ~ 0.50 — half the school-to-school variation is demographics.
- The other half is the school: leadership, instruction, climate, curriculum, principal stability.
- Schools beating their proclivity decile by 20+ pts are doing real work.
- Schools missing their proclivity decile by 20+ pts are failing in a way the citywide ranking obscures.
- Adjusting for proclivity changes which schools look 'good' — wealthy zoned schools fall, mid-poverty over-performers rise.
- This is exactly what NYCENET's comparison group does (just less explicitly).
- Compare with Sean Reardon's 'school-effect' research at Stanford — he isolates ~20-25% of outcome variance to schools specifically.
- What changes if you publish proclivity-adjusted rather than raw rankings?
- Frame: 'Most rankings tell you whose kids the school got. Our adjustment tells you what the school did with them.'
Reporting directions
- Run a regression: composite_es_academic ~ proclivity_score. Report R² + residuals.
- Identify schools where the residual is most positive (punching above) and most negative (failing despite advantages).
- Compare with Sean Reardon's school-effect estimates for NYC.
- Write the explainer + visualization for the methodology.