Low and Pistaferri 2019 — Disability Insurance Error Rates and Gender Differences

Summary

This paper (National Bureau of Economic Research [NBER] Working Paper 26513, 2019; published in the Journal of Political Economy 133(9): 2962–3018, 2025) documents a gender gap in Disability Insurance (DI) Type I errors using Health and Retirement Study (HRS) data (waves 1992–2014) matched to Social Security Administration (SSA) Form 831 administrative records. Women with severe, permanent, work-limiting impairments are ~21.5 percentage points (pp) more likely than observationally equivalent men to have their DI applications rejected — a false rejection rate that arises at the vocational assessment stage (Step 4/5: residual functional capacity [RFC] for other work), not at the medical assessment stage. A structural model with disability vignettes rules out demand-side explanations and identifies SSA supply-side bias as the operative mechanism. Rejected work-limited women do not return to work, confirming these are genuine errors.

Key Claims

• Women with severe, permanent, work-preventing impairments are 21.5 pp more likely to be rejected than observationally equivalent men (Table 3, col. 2; 918 first-round applicants: 357 men, 561 women; sample award rate 37% matches population average).

  • ⚠️ Figure discrepancy (flagged): This page's headline gap of 21.5 pp comes from the narrower Table 3, col. 2 subsample (severe, permanent, work-preventing impairments) — traceable to the NBER working-paper (2019) figures. The published JPE (2025) abstract headlines a 12.8 pp gap for the broader "work-disabled" definition. The two are not necessarily contradictory (different disability definitions/specifications), but the numbers were written from the working paper and have not been reconciled against the published version. Verify against JPE 133(9) before citing a single headline figure.

• The gender gap is entirely at Step 4 (vocational stage): women are 18.7 pp more likely to be rejected on residual functional capacity grounds; there is no gender gap at the medical assessment stage (Table 4).
• Type II error (false award) shows no significant gender gap (Table 3, cols. 3–4).
• Structural model (Section 5): only $\lambda_\mathrm{SSA}$ (SSA threshold bias) is statistically significantly negative; demand-side parameters (underlying health $\lambda_L$, pain threshold $\lambda_L^*$, application cost $\lambda_A^*$) are positive. Demand-side explanations dismissed.
• Vignette evidence (2007 HRS Disability Vignette Survey): respondents rate hypothetical individuals as 1.7 pp less likely to be disabled when the vignette person is female (name randomized, same description), approximating a blind audit of evaluator bias. Women respondents have higher pain thresholds than men respondents.
• Labor market test (Table 10): rejected work-limited women have 2% employment rate in the 3 years post-rejection vs. 19% for rejected women without self-reported limitations; interaction Rejected × Disabled = −0.168 pp (p<0.01) for women vs. −0.020 (not significant, n.s.) for men. SSA's RFC assessment for women is wrong.
• Placebo test (pre-decision employment 5–10 years prior): no differential pre-trends by future disability status, ruling out unobserved heterogeneity.
• Policy implication: gender-blind DI application process (analogous to blind orchestral auditions; Goldin and Rouse 2000) would reduce Type I errors, though some gender-specific disease profiles complicate full blinding.
• Robustness (Table 5): results hold across DI-only sample, different interview-application timing windows, and different disability definitions.

Concepts Introduced or Extended

• DI Classification Errors
• SSA Sequential Determination Process
• Residual Functional Capacity
• DI Denied Population

Entities Mentioned

• Hamish Low
• Luigi Pistaferri
• Health and Retirement Study
• Social Security Administration

Quotes

"We find that women with a severe, work-related, permanent impairment are more likely to have their disability insurance application turned down (i.e., suffer a Type I error) than men with observationally equivalent characteristics."

"These false rejections have lasting consequences. The higher false rejections for women are arising because of an assessment that women have residual work capacity. We show that despite this assessment, these women are not returning to work."

My Take

Methodologically sophisticated: HRS–F831 administrative matching with structural identification via disability vignettes (following Kapteyn, Smith, and van Soest 2007) cleanly separates demand-side from supply-side explanations. The Table 10 labor market outcome test is the paper's strongest result — rejected work-limited women don't work, ruling out the "SSA has superior information" alternative and confirming genuine Type I errors. The gender-blind policy implication is actionable and supported by parallel evidence from orchestral auditions. Limitations: vignette respondents are HRS participants, not professional Disability Determination Services (DDS) evaluators; the paper covers only first-contact decisions and does not track whether gender gaps persist through Administrative Law Judge (ALJ) appeals.