SES explorations summary Updated: 2026-09-10 Scope This synthesis covers the audited 2024-25 elementary-primary SES model family. It uses one official Total Population All Grades row per ordinary school, exact achievement-count identities, student weights based on scored counts, and tested-grade attendance. Middle-school estimates are sensitivities; high and combined-grade schools are excluded from primary tract-SES interpretation. Measures - Adult BA+ and income describe the Census tract containing the school site. They are neighborhood proxies, not direct measures of enrolled families or attendance areas. - SEP describes hardship in the enrolled school population, though it is program-based rather than a complete measure of family resources. - Regular attendance is reconstructed from the assessment-tested-grade rollup. The included-student count validates the rate; scored students weight the achievement model. Main audited findings - ELA: income + BA+ betas 0.209 and 0.577 (R^2 0.557); SEP + BA+ betas -0.649 and 0.273 (R^2 0.747); income + BA+ + attendance betas 0.065, 0.455, and 0.438 (CV R^2 0.677); SEP-model CV R^2 0.768; SEP-model interaction held-out delta 0.000. - Math: income + BA+ betas 0.239 and 0.550 (R^2 0.556); SEP + BA+ betas -0.614 and 0.294 (R^2 0.719); income + BA+ + attendance betas 0.088, 0.430, and 0.446 (CV R^2 0.680); SEP-model CV R^2 0.749; SEP-model interaction held-out delta 0.012. - Science: income + BA+ betas 0.166 and 0.578 (R^2 0.505); SEP + BA+ betas -0.560 and 0.302 (R^2 0.646); income + BA+ + attendance betas 0.110, 0.487, and 0.290 (CV R^2 0.561); SEP-model CV R^2 0.652; SEP-model interaction held-out delta 0.004. Robustness and limits - Repeated half-samples preserve BA+ over income and absolute SEP over BA+ in all three subjects. These are descriptive stability ranges, not inferential confidence intervals. - SEP interval endpoint and exact-only checks do not reverse the principal model ordering. - Pairwise interactions add a modest held-out gain in Math; corrected tests do not support a general interaction improvement across all subjects. - Associations are not causal. Aggregation can hide within-school differences, and a school-site tract may poorly represent schools drawing from wider areas. - Historical and non-core experimental families require their own audits before their older claims are carried into this synthesis. Takeaway The core interpretation is stable but more precise: SEP, adult BA+, and attendance each carry distinct statewide information, while income's additional contribution is comparatively small. The rebuilt pipeline materially improves population validity, denominator integrity, provenance, and reproducibility; finer local interpretations may change more than the statewide pattern.