Why "Poverty" Outpredicts Income in Our Proficiency Models
Generated: 2026-02-21

Short answer
Both statements can be true at once:
- Income and poverty are strongly related.
- Poverty can still be the better predictor of proficiency.

Plain-language explanation
1) Income and poverty overlap, but they are not the same signal.
- In our school-level data, per-capita income and Students Experiencing Poverty are strongly related, but far from identical.
- If correlation is about -0.625, shared variance is only about 0.39, meaning substantial non-overlap remains.

2) The two measures are not equally close to the students being tested.
- Students Experiencing Poverty is a school-population hardship/composition measure.
- ACS per-capita income is a tract-level mean for all residents around a school.
- A measure tied to enrolled students usually tracks school outcomes more directly.

3) Per-capita income is a mean and can be skewed.
- High earners can raise tract mean income even when many enrolled students face hardship.
- That weakens income as a direct proxy for school hardship burden.

4) Poverty captures hardship channels more directly linked to learning disruption.
- ODE's poverty field reflects student hardship categories (for example SNAP/TANF eligibility, foster care, houselessness, migrant status).
- Those conditions can affect attendance continuity, stress load, and day-to-day school access.

5) The model results match this interpretation.
- In non-charter/non-virtual school-level 5-fold CV tests:
  - ELA:
    - BA+ + Attendance + PerCapIncome: 0.5216
    - BA+ + Attendance + Poverty: 0.6534
    - BA+ + Attendance + PerCapIncome + Poverty: 0.6546
  - Math:
    - BA+ + Attendance + PerCapIncome: 0.6444
    - BA+ + Attendance + Poverty: 0.6770
    - BA+ + Attendance + PerCapIncome + Poverty: 0.6768
  - Science:
    - BA+ + Attendance + PerCapIncome: 0.3906
    - BA+ + Attendance + Poverty: 0.5091
    - BA+ + Attendance + PerCapIncome + Poverty: 0.5125

Interpretation for reporting
- Income remains useful as broad community context.
- Poverty appears to carry more independent school-level signal for proficiency.
- Best concise framing:
  "Income and poverty both matter, but poverty is a closer measure of student hardship and therefore a stronger predictor in this model framework."
- Alternate formulation (plain-language):
  "Per-capita income is a neighborhood-wide average for everyone living in the school's tract, while the poverty measure reflects hardship among students actually enrolled in the school."

Caution line (keep with any public use)
- These are school-level observational associations, not causal proof.