Calibrate SSI latent disability and take-up for asset-limit reforms

[MaxGhenis]

Problem

The SSI disability/take-up model belongs in data generation, not only in policyengine-us. Current enhanced CPS inputs make policyengine-us treat broad survey disability as SSI medical eligibility, and takes_up_ssi_if_eligible defaults to true. That makes SSI asset-limit reforms far too large.

Severity

Local 2026 enhanced CPS no-SSI-asset-limit run using policyengine-us 1.700.0 / us-data 1.115.5:

• SSI recipients: 5.797M -> 7.196M, change +1.399M (+24%).
• CBPP no-limit study shared by Kathleen Romig: +462k participants (+6%).
• PolicyEngine adults 18-64: +1.006M vs CBPP +298k.
• PolicyEngine seniors 65+: +393k vs CBPP +115k.
• PolicyEngine children: +0 vs CBPP +49k.
• SSI cash spending: +$12.84B/year.
• Medicaid enrollment: +584k; Medicaid cost: +$20.1B/year.

The current model is about 3x CBPP on new SSI participants, before even considering Medicaid.

Censoring issue

Observed SSI receipt is censored by:

• income eligibility,
• resource eligibility,
• application/take-up,
• SSA medical allowance/denial,
• and possibly state/data reporting.

For people who currently fail income/resources, non-receipt does not tell us whether they would pass SSA disability criteria or apply if the asset rule changed. We need a latent successful-SSI-claim probability, not a deterministic broad-disability flag.

Proposed data-model approach

1. Keep observed current SSI recipients as SSI-success cases.
2. Use CPS variables and other imputed indicators to estimate/assign a probability of successful SSI claim among non-recipients: age, broad disability, work limitation, OASDI/SSDI/disability income indicators, Medicare disability where available, SGA/earnings, household composition, state, current income/resources.
3. Calibrate current baseline SSI recipients/spending to SSA admin totals.
4. Calibrate asset-limit counterfactuals to external evidence, starting with CBPP's +462k no-limit estimate.
5. Export a data variable or take-up/latent eligibility flag that policyengine-us can consume, so relaxing SSI income/assets does not automatically convert every broad-disabled non-recipient into an SSI recipient.

Related

• Calibrate SSI disability and take-up for eligibility-expanding reforms policyengine-us#1043 consumer-side issue.
• Senior and Disabled medicaid needs more details policyengine-us#6010 Medicaid umbrella exposed by this analysis.

Acceptance criteria

• Enhanced CPS contains an input or derived data-backed mechanism for latent SSI successful-claim probability/take-up.
• Existing SSI recipients remain recipients in baseline.
• A no-asset-limit SSI reform produces new participation in the neighborhood of CBPP, with uncertainty documented.
• Children are handled explicitly; current PE result has +0 new child SSI recipients versus CBPP +49k.

[MaxGhenis]

Opened the PolicyEngine-US contract PR that policyengine-us-data should target once released: PolicyEngine/policyengine-us#8364

The proposed variable is meets_ssi_disability_criteria. It is data-overridable and currently falls back to is_disabled & ~ssi_engaged_in_sga; is_ssi_disabled delegates to it. For the US-data follow-up, the imputed SIPP/CPS model should populate this variable directly, then bump policyengine-us to the release that includes it.