Income vs Poverty Reassessment Note
Generated: 2026-02-20

Question
Does the new ODE "Students Experiencing Poverty" field require reassessment of prior assumptions about income and its interactions with attendance/adult education?

Short answer
Yes, a partial reassessment is warranted.
- Income remains useful, but the new poverty field adds substantial independent signal.
- The prior role of income should be interpreted more as a broad community proxy, not the primary hardship measure.
- Income interaction analyses should be retained, but run alongside poverty-aware specifications.

What we tested
- School-level models (non-charter, non-virtual), weighted by participants.
- Outcome: Percent Proficient.
- Core predictors: BA+ rate, attendance, median household income.
- New predictor: Students Experiencing Poverty.
- Compared in-sample R2 and 5-fold cross-validated R2.

Key empirical findings
1) Poverty aligns strongly (but not perfectly) with income context
- Students Experiencing Poverty vs ACS Median HH Income: weighted r about -0.59 to -0.62.
- vs ACS Per Capita Income: weighted r about -0.65 to -0.66.
- Interpretation: strong overlap, but not redundancy.

2) Adding poverty materially improves predictive fit beyond income
Cross-validated R2 (non-charter/non-virtual):
- ELA:
  - BA+ + Attendance + Income: 0.5190
  - BA+ + Attendance + Poverty: 0.6408
  - BA+ + Attendance + Income + Poverty: 0.6508
- Math:
  - BA+ + Attendance + Income: 0.6429
  - BA+ + Attendance + Poverty: 0.6705
  - BA+ + Attendance + Income + Poverty: 0.6726
- Science:
  - BA+ + Attendance + Income: 0.3966
  - BA+ + Attendance + Poverty: 0.5077
  - BA+ + Attendance + Income + Poverty: 0.5195

Interpretation:
- Poverty contributes large incremental signal in ELA and Science, moderate in Math.
- Income still adds some incremental value when poverty is present, but much less than before.

3) Income interactions remain relevant, but poverty-aware interaction models are often stronger
Cross-validated R2 gains from interaction terms:
- Income interactions (income*BA, income*attendance) improve fit in all subjects.
- Poverty interactions (poverty*BA, poverty*attendance) are competitive and in Math outperform income interactions.
- Best fit typically uses both interaction families or whichever is stronger by subject.

Practical implications for our project
1) Do not drop income entirely.
- Keep it as community-resource context.

2) Promote Students Experiencing Poverty to a core factor.
- Treat it as a direct hardship/composition indicator, distinct from tract income.

3) Update interpretation language.
- Prior language implying income as the main socioeconomic channel should be softened.
- Better framing: "Income and poverty both matter; poverty appears to carry additional school-level signal beyond tract income."

4) For interaction analyses and highlights
- Run dual specs:
  - Income-centric interactions (for continuity with prior work), and
  - Poverty-aware interactions (for updated robustness).
- If choosing one headline interaction lens by subject:
  - Math: poverty interactions look especially strong.
  - ELA/Science: both income and poverty interaction sets are useful.

Bottom line
The new field does not invalidate earlier income findings, but it does change emphasis: poverty should now be treated as a first-tier explanatory variable, with income interpreted as a related but broader community proxy.