SES explorations summary Updated: 2026-07-16 Purpose Summarize the current, denominator-verified analyses of socioeconomic context, attendance, and Oregon school achievement. This is a synthesis of several model families, not a claim that one specification identifies causal effects. Scope and conventions - Data: Oregon 2024-25 school-level aggregate achievement rows for English language arts, mathematics, and science. - Student group: Total Population (All Students). - Grade handling: All Grades rows are dropped when the same school has grade-specific rows, avoiding double counting. - Weighting: students with reported Level 1-4 results, using the stored Scored Performance Denominator with a summed Level 1-4 fallback. - Main outcome: Percent Proficient, with Level 1, Level 4, and a four-level achievement index used as sensitivity checks. - The models describe associations among school aggregates. They do not identify effects on individual students or prove that changing a predictor would cause the modeled change in achievement. The measures answer different questions - Adult BA+ and income are Census estimates for the tract containing the school. They describe neighborhood context, not the actual families of every enrolled student. - Students Experiencing Poverty (SEP) is an ODE school-population measure. It more directly describes enrolled students, but it is program-based rather than a complete measure of family resources. - Regular attendance is school-level. It overlaps with SES, while also potentially reflecting school policy and practice. 1) Adult education and income In the two-factor income + adult BA+ models, adult BA+ carries the larger controlled association in every subject: - ELA: beta income 0.090, beta BA+ 0.510, R^2 0.332. - Math: beta income 0.141, beta BA+ 0.497, R^2 0.364. - Science: beta income 0.062, beta BA+ 0.488, R^2 0.283. Across 200 random school-level half-splits per subject, the absolute BA+ beta exceeded the absolute income beta in both halves of every split. This supports a stable ordering in this specification: income is associated with achievement, but adult BA+ retains substantially more independent signal after the two are modeled together. 2) School poverty and adult education Replacing tract income with enrolled-student poverty changes the ordering. In the two-factor SEP + adult BA+ models: - ELA: beta SEP -0.554, beta BA+ 0.216, R^2 0.501. - Math: beta SEP -0.399, beta BA+ 0.330, R^2 0.432. - Science: beta SEP -0.528, beta BA+ 0.192, R^2 0.440. Across 200 half-splits per subject, the absolute SEP beta exceeded the absolute BA+ beta in both halves of every split. A separate ordinary-school two-factor analysis, which excludes charter, virtual, and special-enrollment schools, reaches the same practical conclusion while showing that BA+ still adds information after SEP: - ELA joint R^2 0.596; BA+ adds 0.037 after SEP and SEP adds 0.200 after BA+. - Math joint R^2 0.510; BA+ adds 0.085 and SEP adds 0.095. - Science joint R^2 0.473; BA+ adds 0.031 and SEP adds 0.161. The defensible interpretation is not that one variable has displaced the other. SEP is the stronger enrolled-student hardship signal in these models. Adult BA+ remains a distinct community-context signal, especially in mathematics. 3) Attendance in joint models In the continuity model Percent Proficient ~ income + adult BA+ + attendance, the standardized results are: - ELA: income 0.024, BA+ 0.464, attendance 0.298, R^2 0.417. - Math: income 0.045, BA+ 0.433, attendance 0.412, R^2 0.517. - Science: income 0.005, BA+ 0.459, attendance 0.252, R^2 0.349. The poverty-aware companion models explain more variation: - ELA: SEP + BA+ + attendance R^2 0.537. - Math: SEP + BA+ + attendance R^2 0.545. - Science: SEP + BA+ + attendance R^2 0.469. Ordinary-school cross-validation also favors poverty-aware specifications. Cross-validated R^2 rises from 0.5152 to 0.6516 in ELA, from 0.6394 to 0.6750 in Math, and from 0.3803 to 0.5072 in Science when SEP replaces income alongside BA+ and attendance. Adding income back produces only small further gains: 0.6622, 0.6762, and 0.5188. 4) Interaction and nonlinear checks Pairwise interactions modestly improve the income + BA+ + attendance models: - ELA: Delta R^2 +0.015. - Math: Delta R^2 +0.031. - Science: Delta R^2 +0.031. These increments show some non-additive structure, often involving BA+ and attendance, but they do not overturn the main ordering. An elementary-school nonlinear analysis using all four achievement levels provides a stricter check. BA+ continues to add held-out predictive information beyond a flexible SEP-only curve, particularly for Level 4 and the four-level achievement index. However, a full SEP-by-BA+ interaction surface failed to beat the additive nonlinear model in any of eight held-out outcome comparisons. The evidence supports curved marginal relationships; it does not establish that the BA+ association itself changes reliably with school poverty. 5) Does the proficiency cut create the SES result? The main context pattern survives when Percent Proficient is replaced with a four-level index constructed from Level 1-4 counts: - ELA: Percent Proficient correlations are BA+ 0.65, SEP -0.75, attendance 0.57; index correlations are 0.64, -0.75, and 0.59. - Math: Percent Proficient correlations are 0.66, -0.65, and 0.69; index correlations are 0.65, -0.64, and 0.70. - Science: Percent Proficient correlations are 0.58, -0.67, and 0.45; index correlations are 0.54, -0.63, and 0.53. That is reassuring for broad context claims, but not a license to treat proficiency as a complete score distribution. Level 1, Level 4, and the middle levels can move differently. Tail-specific claims and school rankings should be checked against multiple level-based outcomes. 6) Geography and measurement quality The statewide association is not equally sharp in every setting. Adult BA+ models are stronger in suburban and city schools than in town and rural schools. Larger schools also produce stronger fits than smaller schools. These differences can reflect true context differences, school-boundary mismatch, enrollment choice, tract geometry, and ordinary small-sample variation. For that reason, tract measures should be treated as neighborhood proxies rather than descriptions of each school's enrolled families. Elementary-only and geography-aware checks are especially important when analyzing individual residual schools. 7) Spending and class size in context The current school-level cross-sections do not show a clean class-size contribution comparable to SEP, BA+, or attendance after controls. Spending has detectable but inconsistent secondary signal and is sensitive to specification and collinearity. These findings do not establish that resources have no effect; the available single-year aggregate data are poorly suited to estimating causal resource effects. Overall takeaway - No single variable is a sufficient summary of socioeconomic context. - SEP is the strongest and most stable enrolled-student hardship measure in the current 2024-25 models. - Adult BA+ is the most stable tract-based context measure and retains independent signal after SEP is included. - Attendance contributes substantial additional information, especially in mathematics. - Income remains useful descriptive context but contributes little unique signal after BA+ and attendance in the continuity model. - The broad SES-achievement relationship survives split samples, alternate weighting perspectives, and achievement-level sensitivity checks. - The results remain descriptive, geographically imperfect, and unsuitable for deterministic judgments about individual schools. Related Evidence Lab materials - Adult BA+ and Student Poverty: Two-Factor Check. - SES + Attendance Joint Model Report. - Income, Education, and Attendance Interaction Report. - Income and Poverty Reassessment Note. - Why School Poverty Can Outpredict Income. - BA+ Signal in High-Poverty Schools. - Ho Proficiency Thresholds: Oregon 2024-25 Exploratory Read. - Spending and Class Size vs Performance: Findings Memo.