Carry educational attainment as an auxiliary variable (imputation predictor + ACS-calibratable margin)

Proposal

Carry educational attainment (ASEC A_HGA, collapsed to a small class set, e.g. the SCF's four edcl classes) on the populace-us support spine as an auxiliary variable: not a rules-engine input, but a conditioning variable for imputation and a calibratable margin.

Why

Imputation quality. Attainment is among the strongest available predictors of income and wealth conditionals. Standard multiple-imputation guidance is to include strong auxiliaries in the imputation model even when no downstream analysis consumes them — the joint quality of the variables we do care about depends on conditioning richness. The imputation-paper experiments (github.com/PolicyEngine/imputation-paper) had to drop education from the SCF↔CPS shared predictor set and its absence visibly weakens the conditional signal; its "populace-scale" experiment adds it back from raw ASEC and measures the effect.
Calibration surface. ACS publishes attainment margins by age and state (S1501), so the variable is not just a predictor: it is a target family for the local-area program, cheap to add to the Ledger lane.
Cost is low. A_HGA ships in every ASEC vintage the spine pools; the mapping to a 4-class attainment variable is a few lines in unit construction.

Scope note

Flag it in provenance as auxiliary (carried for conditioning/calibration, not consumed by rules), so nobody reads its presence as a tax-benefit modeling claim. Related: #253 (tuition expenses) is about a rules input and is orthogonal — this issue is about attainment as a conditioning/calibration variable.

🤖 Generated with Claude Code

This is now a deliverable for the student debt pilot (tracker: PolicyEngine/us-student-debt#1). Appendix A month 1, Project 2: "Add educational attainment to Microcosm and calibrate against Census state totals." Month 3 then computes "state-level degree ROI before/after taxes, transfers, and net of debt payments" for a two-year and a four-year degree. Assigning to David.

Changes to the proposal above, needed for that use:

Classes. The SCF's four edcl classes can't separate an associate's degree from "some college". Keep at least: less than high school, high school diploma or GED, some college with no degree, associate's (occupational and academic together), bachelor's, and graduate or professional. ASEC A_HGA has these codes, and so does ACS SCHL for the ACS-local lane.
Readable by the analysis. The ROI work reruns PolicyEngine with counterfactual earnings for each degree-holder, so the attainment column has to sit next to PE outputs for the same person. Either add a policyengine-us input enum (named for its source, e.g. cps_educational_attainment, per the explicit-definition-names rule), or export it as a Microcosm auxiliary column the analysis joins on person id. Pick one and note it here.
Calibration. Use state totals from ACS 1-year S1501. The Chronicle package is in PolicyEngine/chronicle#310. Also check the SCF debt imputation (PolicyEngine/microcosm#1073) with attainment as a predictor, since education debt depends heavily on it.
Timing. Month 1 ends about 10/21. The client's causal multipliers arrive by the end of month 2 (their obligation; Addendum 1 clause 5). Attainment should be in a build before then.

Acceptance: the column is in a US build; state totals are compared with S1501 (receipt); the column is flagged in provenance as described above; and there's a short note on which classes the ROI analysis uses.

Carry CPS educational attainment as an auxiliary column and bind S150…

Point the #297 runner at the data shard's stored-input tests

Add the #297 receipt and its findings on instrument differences and t…

Quote the merged some-college row's national gap in the #297 receipt

Carry CPS educational attainment as an auxiliary column and bind S150…

Choice: a Microcosm auxiliary column, not a policyengine-us input. The fiscal-refresh build writes cps_educational_attainment (and the raw A_HGA beside it) on the person table with --cps-educational-attainment; the ROI analysis joins it on person_id. No policyengine-us change is needed, and the stored-input register flags it as auxiliary (US_AUXILIARY_STORED_COLUMNS: carried for conditioning and calibration, not consumed by rules).

Classes (the seven S1501 population-25+ categories; the first two merge to the six asked for):

Class	ASEC `A_HGA`	ACS `SCHL`
`less_than_9th_grade`	31-34	01-11
`9th_to_12th_grade_no_diploma`	35-38	12-15
`high_school_graduate_includes_equivalency`	39	16-17
`some_college_no_degree`	40	18-19
`associates_degree` (occupational + academic)	41-42	20
`bachelors_degree`	43	21
`graduate_or_professional_degree`	44-46	22-24
`not_in_universe`	0 (under 15)	blank, or under 15

Codes checked against the ASEC 2023-2026 public-use data dictionaries (Item 18h) and the 2024 PUMS dictionary. The ROI analysis uses associates_degree (two-year) and bachelors_degree (four-year) against the high_school_graduate_includes_equivalency baseline.

Draft PR: #1356

Record the #297 receipt at the branch head (Modal run 20261022T171517…

Navigation Menu

Repository navigation

Carry educational attainment as an auxiliary variable (imputation predictor + ACS-calibratable margin) #297

Proposal

Why

Scope note

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Carry educational attainment as an auxiliary variable (imputation predictor + ACS-calibratable margin) #297

Description

Proposal

Why

Scope note

Activity

MaxGhenis commented on Sep 30, 2026

DTrim99 commented on Oct 22, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions