# Oregon BA+ and School Poverty Joint Model Check

Date: 2026-05-19

## Question

Have we tested whether ACS adult BA+ still predicts proficiency after adding the school-sourced `Students Experiencing Poverty` measure to the same model?

Short answer: earlier Oregon work did something very close in broader four-factor and residual-peer models. This artifact runs the narrower two-predictor diagnostic directly:

`Percent Proficient ~ ACS BA+ + Students Experiencing Poverty`

## Scope

- Oregon 2024-25 ELA, Math, and Science processed school files.
- Student group: `Total Population (All Students)`.
- Primary filter: non-charter, non-virtual schools.
- Sensitivity filter: all schools.
- School-level aggregation: grade rows are aggregated to one school result within each grade band; dataset-backed `All Grades` rows are not double-counted when grade rows exist.
- Weighting: `Number of Participants`.
- Outcome: percent proficient.
- Predictors: ACS tract/block-group adult BA+ rate and ODE school-sourced `Students Experiencing Poverty`.

## Primary Results: All Tested Grades

| Subject | Schools | BA+ r | Poverty r | Joint R^2 | BA+ beta | Poverty beta | BA+ delta R^2 after poverty | Poverty delta R^2 after BA+ |
| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| ELA | 1025 | 0.623 | -0.751 | 0.597 | 0.235 | -0.599 | 0.032 | 0.209 |
| Math | 999 | 0.641 | -0.658 | 0.513 | 0.370 | -0.419 | 0.080 | 0.102 |
| Science | 990 | 0.555 | -0.668 | 0.475 | 0.219 | -0.529 | 0.029 | 0.167 |

## Primary Results: Elementary Scope

| Subject | Schools | BA+ r | Poverty r | Joint R^2 | BA+ beta | Poverty beta | BA+ delta R^2 after poverty | Poverty delta R^2 after BA+ |
| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| ELA | 621 | 0.674 | -0.805 | 0.682 | 0.250 | -0.639 | 0.035 | 0.229 |
| Math | 621 | 0.665 | -0.779 | 0.646 | 0.266 | -0.602 | 0.040 | 0.203 |
| Science | 584 | 0.646 | -0.735 | 0.584 | 0.282 | -0.547 | 0.044 | 0.167 |

## Read

School poverty is very strong, especially for ELA and Science, but BA+ generally remains positive with meaningful incremental explanatory power after school poverty is included.

If the school-sourced poverty measure captured nearly all of the relevant socioeconomic context, then the BA+ standardized beta and drop-one delta R^2 would fall near zero. In these Oregon models, they do not.

That does not make the ACS school-site measure perfect. It still says Oregon's tract BA+ measure is capturing context that is not fully reducible to the school poverty field. The school poverty field also captures a lot that BA+ does not, so the best current interpretation is complementary signal rather than replacement.

## Caveats

- This is observational and school-level. It does not identify individual family SES effects.
- `Students Experiencing Poverty` is a school-sourced Oregon field and should not be read as an individual-student poverty measure.
- BA+ is measured from ACS geography attached to the school location, not a true attendance-zone crosswalk.
- High school and mixed-grade results remain more vulnerable to catchment-area mismatch than elementary results.
- Standardized betas compare variables on common weighted standard-deviation units; delta R^2 is the added explanatory power from adding one predictor after the other.

## Output

- `artifacts/reports/oregon_ba_school_poverty_joint_model_2026-05-19.csv`