Why "Poverty" Outpredicts Income in Our Proficiency Models

Why "Poverty" Outpredicts Income in Our Proficiency Models
Regenerated: 2026-09-10

Short answer
Income and poverty overlap substantially, but they measure different populations. In these audited elementary-primary models, ODE Students Experiencing Poverty is more closely associated with school proficiency and predicts held-out school results better than school-site tract income.

What the measures represent
- SEP is an ODE school-population hardship measure tied to enrolled students.
- Median household and per-capita income are ACS estimates for all residents of the tract containing the school. They describe the surrounding neighborhood, not the families of all enrolled students.
- Differences can reflect enrollment boundaries, school choice, tract composition, and the fact that a tract average and a student hardship measure are not interchangeable.

Observed overlap
- ELA: SEP correlation with median household income -0.735 and with per-capita income -0.711.
- Math: SEP correlation with median household income -0.736 and with per-capita income -0.711.
- Science: SEP correlation with median household income -0.739 and with per-capita income -0.717.
These correlations imply roughly 54% shared variance with median household income and about 51% with per-capita income. Substantial overlap therefore coexists with substantial non-overlap.

Held-out predictive comparison
Mean repeated five-fold CV R^2; every model also includes adult BA+ and tested-grade attendance:
- ELA: income 0.6767; poverty 0.7676; income + poverty 0.7727.
- Math: income 0.6788; poverty 0.7475; income + poverty 0.7485.
- Science: income 0.5636; poverty 0.6538; income + poverty 0.6546.
The same ordering survives lower-endpoint, midpoint, upper-endpoint, and exact-only treatment of SEP's few left-censored values.

Interpretation
Poverty was the stronger predictor in these elementary-primary specifications. A plausible reason is that SEP is measured on the enrolled population while tract income is a broader neighborhood proxy, but these observational models do not identify the mechanism or prove a causal effect.

Plain-language formulation
"Income describes the neighborhood around a school. The poverty measure is tied more directly to students enrolled there, and in these models it predicts school results better."