qualified_tuition_expenses essentially unimputed (641 records, $4.5B) — education credits understated ≥40%

Summary

Education tax credits validate 40% below IRS actuals, and the mechanism is a near-empty input: qualified_tuition_expenses is populated on only 641 of 160,858 person records (weighted $4.5B) in the released dataset.

Evidence

On populace_us_2024.h5 (…f0af251…20260620):

For scale: roughly 8–9M returns claim education credits (SOI TY2023: 7.2M nonrefundable-credit returns), against ~19M postsecondary students; aggregate qualified tuition paid is tens of billions. $4.5B across 641 records means the input is essentially absent — the simulated credit can only reach 60% of actual claims, and the refundable AOTC and any tuition-linked analysis inherit the same gap.

Suggested fix

Impute qualified_tuition_expenses (and the AOTC eligibility inputs — enrollment status/half-time flags) from enrollment + published tuition distributions (NCES/IPEDS), or calibrate education-credit claims to the SOI totals.

Found via the SOI-actuals baseline backtest on the reform-validation page (PolicyEngine/calibration-diagnostics). Sibling of #225 (SSN) and #252 (auto-loan) — same class: model logic exists, input never imputed.

Start with the reform-validation page and PolicyEngine/calibration-diagnostics backtest that exposed the gap, then inspect how qualified_tuition_expenses and enrollment or half-time inputs are produced for populace_us_2024.h5. Review the NCES/IPEDS and SOI references in the issue; done means education-credit inputs are no longer effectively absent and simulated claims are calibrated against the cited SOI totals.