Mitchell and Thompson 2025 — Differences in Disability Insurance Allowance Rates by Race and Ethnicity

Summary

Mitchell and Thompson link the 2015 American Community Survey (ACS) to Social Security Administration (SSA) administrative records (831 file, Detailed Earnings Record [DER], Numident) to estimate racial/ethnic differences in disability insurance (DI) allowance rates at the initial and reconsideration stages for Title II applicants aged 25–65. Large raw allowance gaps — up to 13.3 percentage points (pp) lower for non-Hispanic (NH) American Indian/Alaska Native (AIAN) males relative to NH White — collapse to statistical insignificance for 7 of 8 race × sex comparisons after adjusting for age, body system code, pre-disability earnings, education, marital status, and region using three separate methods (propensity score matching [PSM], inverse probability weighting [IPW], logistic regression with G-computation). A small residual gap of roughly 1.7–2.1 pp persists for NH Black males under two of the three estimators but not the third, suggesting the evidence for a residual adjudication disparity is fragile and method-sensitive. The paper is silent on Administrative Law Judge (ALJ)-stage hearing decisions, where prior work has found unexplained racial gaps.

Key Claims

• Raw allowance rate gaps vs. NH White males: NH AIAN –13.3 pp; NH Black –9.5 pp; Hispanic –5.4 pp; NH Asian +3.1 pp. For females: NH AIAN –12.1 pp; NH Black –8.1 pp; Hispanic –4.7 pp; NH Asian +0.7 pp.
• After adjustment for observable characteristics, 7 of 8 group × sex raw gaps become statistically indistinguishable from zero across all three estimation methods.
• Residual NH Black male gap: –1.7 pp (logistic G-computation) to –2.1 pp (IPW), statistically significant. Not significant under PSM (–0.4 pp). Residual for NH Black females: small (–0.5 to –0.9 pp), not significant.
• Dominant predictors of allowance: body system code (genito-urinary odds ratio [OR] ~11; malignant neoplasms OR ~10), age at application 55–65 (OR ~5.8), pre-disability earnings — all orders of magnitude larger than any race/ethnicity odds ratio.
• Application propensity paradox: NH Black applies ~70% more often than NH White despite reporting disability only ~30% more often; NH AIAN applies ~40% more despite ~70% higher disability prevalence. These disparities drive the raw allowance gaps because low-propensity applicants are on average closer to the medical eligibility margin.
• Race data quality problem solved: SSA's administrative race variable has selection bias (enumeration at birth since 1987; coding shifts from 3 to 5 categories around 1980). ACS linkage (94.4% Protected Identification Key [PIK] match rate) provides Office of Management and Budget (OMB)-standard race classification and solves this.
• Three estimators converge on the same substantive conclusion for 7 of 8 groups, supporting robustness. The NH Black male residual is the only point of non-convergence.
• Propensity score match rates: >90% for NH Black, NH Asian, Hispanic; NH AIAN 88.1% (male) / 87.5% (female). Covariate balance standardized mean difference (SMD) <0.1 in all but one case (NH AIAN male: 0.103 on endocrine code).

Concepts Introduced or Extended

• SSA Sequential Determination Process
• DI Application Costs and Take-Up
• DI Denied Population

Entities Mentioned

• Joshua Mitchell
• Daniel Thompson

Quotes

"After controlling for these observable characteristics, the allowance rate differences between non-Hispanic White applicants and those of other race/ethnicity groups are generally small and statistically indistinguishable from zero."

"These findings suggest that conditional on having applied for DI benefits, differences in allowance rates across racial/ethnic groups at the initial and reconsideration stages appear to be largely explained by observable characteristics of applicants."

My Take

The paper makes a genuine methodological contribution by solving the SSA race data problem through ACS linkage and by deploying three independent estimators to test robustness. The key substantive finding — that body system code, age, and earnings dominate race in predicting allowance — is important and policy-relevant. Two caveats deserve emphasis. First, the paper covers only Disability Determination Services (DDS) initial + reconsideration; the Government Accountability Office (GAO 1992, 2004) finding of unexplained racial gaps at the ALJ hearing stage is the more politically salient concern, and this design cannot address it. Second, the application propensity paradox is perhaps the paper's most striking finding: Black applicants apply at far higher rates relative to their self-reported disability than White applicants, suggesting meaningful selection into the applicant pool by race — an access or behavioral barrier upstream of adjudication that no amount of adjudication fairness can fix. The fragility of the NH Black male residual gap (significant under 2 of 3 methods) warrants cautious interpretation: there may be a small residual adjudication disadvantage, but the data cannot establish it firmly.