Statistical findings summary

Why numeric analysis was needed

Dashboard charts and maps reveal broad patterns, but key factors are correlated with each other. Income, adult education, attendance, and school-level poverty often move together, so visual inspection alone cannot cleanly separate their contributions.

Core methodology

• Weighted correlations and weighted regression models, using participant counts.
• Joint models to compare relative signal strength under controls.
• Dual-spec checks: income-based continuity models and poverty-aware companion models.
• Interaction checks to test whether one factor's association changes at different levels of another.
• Split-sample stability checks to verify findings are not driven by a narrow subset.
• Focused exploratory models for spending and class size under collinearity-aware settings.
• For detailed numeric outputs by script, see the statistical results overview.

Main findings to date

• The strongest school-context signals are student poverty concentration, attendance, and adult BA+ context. Which one leads depends on subject and model specification.
• Attendance adds substantial explanatory power and remains a top-tier factor after SES controls.
• ODE Students Experiencing Poverty is a first-tier factor in 2024-25 models and often rivals or exceeds adult BA+ in explanatory strength.
• Income remains useful as a community-context measure, but is usually smaller after controls.
• Funding and class-size signals are weaker and less stable than education/attendance in current cross-sectional views.
• Some interaction terms (especially education x attendance) improve fit, suggesting conditional effects.

What appears robust

• Education-over-income ordering persisted in repeated split checks and in historical comparisons performed so far.
• In 2024-25 analyses, poverty-aware models generally outperform income-only analogs without displacing BA+ leadership.
• Attendance remains a major co-varying factor and should be interpreted alongside SES, not as a side note.

Current limits

• Analyses are school-level aggregates; they do not identify individual-level causal pathways.
• Cross-sectional design limits causal interpretation, especially for spending/class-size effects.
• Collinearity between spending measures requires caution when reading raw coefficient signs.
• Poverty-aware findings currently apply to 2024-25 models; 2018-19 comparisons remain income/education-based.

Detailed numeric overview

Read a fuller, model-by-model summary with effect sizes, fit improvements, and interpretation notes.

Open statistical results overview

Scripts and reports

Download the analysis scripts and summary reports used for the findings above. These files are provided so readers can inspect methods and replicate results.

• Joint SES model script
• SES + attendance joint model script
• Interaction analysis script
• Split-stability script
• Spending/class-size ridge script
• SES explorations summary report (HTML)
• Statistical results overview summary page (HTML)
• Statistical results overview full 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)