Oregon School Data Explorer

Income vs Poverty Reassessment Note

Published analysis report from Evidence Lab artifacts.

Income vs Poverty Reassessment Note
Corrected purpose-sensitive rebuild: 2026-09-10.
Original artifact date retained in filename: 2026-02-20.

Question
How do enrolled-student poverty and school-address tract income compare for ordinary neighborhood elementary schools when tract adult BA+ and tested-grade attendance are held consistent?

Primary estimand and units
- Academic year: 2024-2025; population: Total Population (All Students); outcome row: official ODE All Grades.
- Primary universe: explicit ODE Regular schools, excluding charter, virtual, and curated special-enrollment flags, with a nonempty cross-subject observed tested-grade set confined to grades 3-5.
- Middle schools with observed tested grades confined to 6-8 are reported only as a sensitivity. Mixed-grade and high-school profiles are excluded.
- Outcome is proficiency percentage points, calculated as 100 * (Level 3 + Level 4) / Scored Performance Denominator.
- Every retained row must have complete integer Level 1-4 counts, their sum must equal Scored Performance Denominator, and Level 3 + Level 4 must equal Number Proficient.
- Model weights are scored students. Participant counts are not achievement denominators.
- BA+ is a tract adult share converted from a 0-1 fraction to percent; attendance and SEP are 0-100 percent; income fields are nominal ACS dollars.
- Primary SEP models use ODE exact values plus the ETL's documented interval midpoint for valid left-censored values (2.5 percentage points for <5%). These midpoint values are explicit estimates, not exact observations.
- Lower- and upper-endpoint scenarios test the disclosed interval bounds; an exact-only complete-case scenario tests sensitivity to dropping censored schools.
- CV R^2 is five-fold out-of-sample prediction repeated over five fixed school-row splits. In-sample weighted R^2 is separately labeled.

Primary elementary documented-midpoint SEP results
- ELA: model-specific samples income=457, poverty=457, both=457 schools; income-model scored students=84,996; mean repeated-CV weighted R^2 income=0.6767, poverty=0.7676, both=0.7727; poverty-minus-income=+0.0909, both-minus-income=+0.0961.
- Math: model-specific samples income=456, poverty=456, both=456 schools; income-model scored students=84,607; mean repeated-CV weighted R^2 income=0.6788, poverty=0.7475, both=0.7485; poverty-minus-income=+0.0687, both-minus-income=+0.0697.
- Science: model-specific samples income=433, poverty=433, both=433 schools; income-model scored students=28,388; mean repeated-CV weighted R^2 income=0.5636, poverty=0.6538, both=0.6546; poverty-minus-income=+0.0902, both-minus-income=+0.0910.
- Sensitivity conclusion: for every subject, lower endpoint, midpoint, upper endpoint, and exact-only complete cases preserve the same CV R^2 ordering (income < poverty < income plus poverty); the SEP treatment does not change the substantive ranking.

Coverage and sensitivity
- ELA elementary (primary): 457 validated BA/attendance rows before estimand-specific missingness; 452 exact-SEP rows; 457 rows in the documented-midpoint scenario; 13 rows omitted for unavailable or invalid attendance; 0 rows failed achievement-count validation.
- ELA middle (sensitivity): 188 validated BA/attendance rows before estimand-specific missingness; 188 exact-SEP rows; 188 rows in the documented-midpoint scenario; 5 rows omitted for unavailable or invalid attendance; 0 rows failed achievement-count validation.
- Math elementary (primary): 456 validated BA/attendance rows before estimand-specific missingness; 451 exact-SEP rows; 456 rows in the documented-midpoint scenario; 13 rows omitted for unavailable or invalid attendance; 0 rows failed achievement-count validation.
- Math middle (sensitivity): 187 validated BA/attendance rows before estimand-specific missingness; 187 exact-SEP rows; 187 rows in the documented-midpoint scenario; 5 rows omitted for unavailable or invalid attendance; 0 rows failed achievement-count validation.
- Science elementary (primary): 433 validated BA/attendance rows before estimand-specific missingness; 427 exact-SEP rows; 433 rows in the documented-midpoint scenario; 5 rows omitted for unavailable or invalid attendance; 0 rows failed achievement-count validation.
- Science middle (sensitivity): 188 validated BA/attendance rows before estimand-specific missingness; 188 exact-SEP rows; 188 rows in the documented-midpoint scenario; 2 rows omitted for unavailable or invalid attendance; 0 rows failed achievement-count validation.

Interpretation
- The numeric contrasts above are derived from the machine-readable model table, not manually transcribed claims.
- Tract SES is treated as school-site neighborhood context. The elementary restriction makes that proxy more defensible; middle-school estimates remain labeled sensitivities because their catchments commonly exceed one tract.
- The documented midpoint defines the primary SEP estimate so the lowest-poverty censored schools are not dropped from the headline sample. Lower/upper endpoints and exact-only complete cases remain separate sensitivities. Each model and correlation applies only its own required-field missingness rule.
- These are descriptive school-level associations and predictive comparisons, not causal estimates or student-level relationships.

Machine-readable companions
- income_poverty_reassessment_model_summary_2026-09-10.csv: model metrics, explicit units, in-sample fit, repeated-CV fit, and correlations.
- income_poverty_reassessment_coverage_2026-09-10.csv: filter, validation, and SEP-scenario coverage counts.
- income_poverty_reassessment_school_rows_2026-09-10.csv: validated elementary and middle school rows before SEP-scenario selection.
Launch dashboard