Statistical results overview

• The strongest recurring school-level signals are student poverty concentration, attendance, and adult BA+ context.
• Income remains correlated with outcomes, but usually has small unique signal after controls.
• In 2024-25 models, ODE Students Experiencing Poverty is a first-tier factor and often rivals or exceeds adult BA+ in explanatory strength.
• Spending/class-size effects are weaker and less stable in statewide cross-sectional models.
• Core ordering remains stable across split checks and pre-pandemic (2018-2019 Math) runs.

• Unit: school rows (not student-level microdata).
• Weights: participant counts.
• Main outcome: Percent Proficient.
• Baseline continuity spec: income, adult BA+ rate, attendance.
• Companion 2024-25 spec: poverty, adult BA+ rate, attendance (with optional income add-back).
• Exploratory predictors: overall spending per student, classroom spending per student, median class size.

Baseline continuity spec: Percent Proficient ~ income + education + attendance

Education and attendance dominate the controlled signal in all three subjects.

In non-charter/non-virtual cross-validated tests, replacing income with Students Experiencing Poverty materially improved fit in all three subjects.

• ELA CV R^2: 0.5190 (income spec) vs 0.6408 (poverty spec); with both: 0.6508.
• Math CV R^2: 0.6429 (income spec) vs 0.6705 (poverty spec); with both: 0.6726.
• Science CV R^2: 0.3966 (income spec) vs 0.5077 (poverty spec); with both: 0.5195.

Interpretation: poverty appears to be a closer school-population hardship signal, while income remains useful community context. These values use the median-household-income continuity spec; the poverty explainer includes a per-capita-income variant with slightly different baselines.

Interaction terms improved fit in every subject. The largest incremental gains were typically tied to education x attendance, indicating the strength of one factor can change by the level of the other.

• ELA: interaction model Delta R^2 = +0.015
• Math: interaction model Delta R^2 = +0.031
• Science: interaction model Delta R^2 = +0.031

For high-end outcomes (for example, Math Percent Level 4), education can become even more dominant. In one representative run, Level 4 had beta income = 0.133 vs beta education = 0.533 (R^2 = 0.399).

Spending variables are highly collinear with each other and with SES context. Ridge and permutation-based checks showed weaker and less stable contributions than education/attendance. Median class size was often near-zero after controls.

• Split-sample checks preserve education > income ordering.
• 2024-25 poverty-aware checks preserve BA+ prominence while improving model fit.
• Pre-pandemic 2018-2019 Math analyses (with era-appropriate ACS) preserve the same ordering.
• Magnitude shifts with scope filters, but ordering is stable.

• These are association models, not causal estimates.
• School-level aggregation can hide within-school heterogeneity.
• Use consistency across model families and split checks as the main robustness lens.

• Statistical results overview full report (HTML)
• SES explorations summary report (HTML)
• SES + attendance joint model report (HTML)
• Interaction report (HTML)
• Spending/class-size findings memo (HTML)
• Income vs poverty reassessment note (HTML)
• BA+ signal in high-poverty schools (HTML)
• Why poverty can outpredict income (HTML)