Income vs Poverty Reassessment Note

Question
How do school poverty and tract income compare when adult BA+, attendance, and ordinary-school scope are held consistent?

Scope
- School-level Total Population models for Oregon 2024-25.
- Charter, virtual, and curated special-enrollment schools excluded.
- Outcomes and weights use students with reported Level 1-4 results.
- Five-fold out-of-sample predictions use a fixed random seed for reproducibility.
- Percent Proficient is calculated from reported Level 3 and Level 4 counts.

Main findings
- Income and poverty overlap substantially but are not interchangeable.
- School poverty adds substantial predictive information in ELA and Science and a smaller increment in Math.
- Median household income contributes some additional information when school poverty is present, but much less than poverty contributes to the income model.

Weighted correlations: Students Experiencing Poverty vs tract income
- ELA: median household income r=-0.621; per-capita income r=-0.659; 1,017 schools; 236,225 students with reported Level 1-4 results.
- Math: median household income r=-0.622; per-capita income r=-0.661; 992 schools; 230,300 students with reported Level 1-4 results.
- Science: median household income r=-0.596; per-capita income r=-0.649; 984 schools; 99,887 students with reported Level 1-4 results.

Five-fold cross-validated R^2
- ELA:
- BA+ + Attendance + Income: 0.5152
- BA+ + Attendance + Poverty: 0.6516
- BA+ + Attendance + Income + Poverty: 0.6622
- Math:
- BA+ + Attendance + Income: 0.6394
- BA+ + Attendance + Poverty: 0.6750
- BA+ + Attendance + Income + Poverty: 0.6762
- Science:
- BA+ + Attendance + Income: 0.3803
- BA+ + Attendance + Poverty: 0.5072
- BA+ + Attendance + Income + Poverty: 0.5188

Interpretation
- ODE Students Experiencing Poverty is the more direct enrolled-student hardship measure.
- Tract income remains useful as broader community context and adds a modest increment in some specifications.
- Adult BA+ and attendance remain in every model so the income-versus-poverty comparison does not mistake their shared signal for a poverty or income effect.
- These are descriptive school-level associations, not causal estimates.