Summary
Maestas, Mullen, and Strand present the first causal estimates of Social Security Disability Insurance (SSDI)'s work-disincentive effect using the full applicant population, exploiting quasi-random assignment of applications to Disability Determination Services (DDS) examiners as an instrument for benefit receipt. Using the universe of 2005–2006 SSDI initial determinations (2.2 million applications, ~8,700 examiners), linked to earnings records and ultimate benefit-receipt data through 2010, they find that employment of marginal entrants would be 28 percentage points (pp) higher two years after the initial decision had they not received benefits — roughly half the ordinary least squares (OLS) estimate, the remainder explained by selection of more severely impaired individuals into SSDI. The 28 pp local average treatment effect (LATE) applies only to the 23% of applicants who are on the margin of program entry; the majority are either clearly eligible (always takers) or clearly ineligible (never takers), and within the marginal group the effect is highly heterogeneous, ranging from statistically zero at the high-severity end to over 50 pp at the low-severity end.
Key Claims
- LATE = 28 pp employment (2 years post-decision, 2005 cohort). OLS gives ~35 pp — overstated because more severely impaired individuals select into SSDI. Instrumental variables (IV) corrects for this selection. Above–substantial gainful activity (SGA) employment: ~18–19 pp higher; annual earnings: ~$3,800 higher (including zeros). Effects fade to 16–21 pp at 3–4 years.
- Instrument: each examiner's leave-one-out allowance rate (EXALLOW). First stage: a 10 pp increase in examiner leniency → ~2.3 pp increase in ultimate SSDI receipt (t ≈ 40). Identifying assumptions: conditional random assignment within DDS (supported by stability of coefficient as covariates added) and monotonicity (supported by positive cross-impairment correlation of examiner rates and increasing slope of SSDI receipt as a function of EXALLOW).
- 23% marginal: Only 23% of applicants are on the margin of program entry. 57% are "always takers" (would receive SSDI regardless of examiner), 20% are "never takers." The marginal group falls between the 20th and 43rd percentiles of the unobserved severity distribution.
- Marginal entrant profile: 43% more likely than average to have a mental disorder; 30% less likely to have musculoskeletal impairments; 43% more likely to be very young (<29) or near retirement age (60–64); more likely to have pre-onset earnings in the lowest quintile. Not a typical "borderline" musculoskeletal case — the mental disorder result is the most striking departure from the conventional view.
- Marginal treatment effect (MTE) heterogeneity: Marginal Treatment Effects decline monotonically with unobserved severity. At the high-severity end of the marginal group, MTE ≈ 0 (statistically zero). At the low-severity end, MTE > 50 pp. Implication: policies that expand access to SSDI add high-disincentive cases at the margin; policies restricting access remove them.
- Impairment-type heterogeneity: Mental disorder applicants face a 36.6 pp employment effect vs. only 17.3 pp for musculoskeletal applicants. Younger (30–44) applicants: ~34 pp; ages 60–64: ~10.5 pp. Bottom earnings quintile: 29.6 pp.
- Processing-time bias: Strict examiners' denied cases face ~125 days of additional appeals processing. Skill depreciation at ~0.4 pp/month during appeals means the 28 pp IV estimate may be slightly understated — a back-of-envelope correction yields ~35–36 pp, close to or slightly above the OLS upper bound. This resurrects Parsons (1991): the OLS observational upper bound is not valid if the application/appeals process itself depresses labor supply.
Concepts Introduced or Extended
- Causal Effects of DI Receipt
- Disability Insurance Program — first causal (IV) estimate of the SSDI work disincentive; 23% marginal population; MTE heterogeneity as input to program design
- SSA Sequential Determination Process — examiner assignment exploits the DDS adjudication system; the five-step process and appeals structure are integral to the identification strategy
- DI Denied Population — the 23% marginal population overlaps substantially with the denial-margin population; Maestas et al. provide causal estimates of what the marginal denials would have done in the counterfactual
- Substantial Gainful Activity — SGA is both a Step 1 screen in the determination process and the above-SGA earnings outcome measure ($11,760/year in 2009)
- Conditional DI Applicants — the marginal entrant's profile (young, mental disorder, low earnings) partially overlaps with the conditional applicant concept, though the mechanisms are distinct: marginal entrants are identified by examiner leniency, not by recession-era labor force exit
Entities Mentioned
Quotes
"We find that among the estimated 23 percent of applicants on the margin of program entry, employment would have been 28 percentage points higher had they not received benefits."
"The effect is heterogeneous, ranging from no effect for those with more severe impairments to 50 percentage points for entrants with relatively less severe impairments."
My Take
The examiner-assignment design is one of the cleanest identification strategies in the DI literature and the key reason this paper became canonical. The 28 pp LATE is substantially smaller than the 35 pp OLS estimate — demonstrating that naive comparisons of allowed vs. denied applicants overstate the disincentive by ~25% because of severity selection. The more important finding is probably the heterogeneity: two-thirds of the applicant population (always takers) are so clearly eligible that examiner leniency barely affects them; the remaining marginal population has a disincentive effect that varies from zero to >50 pp, which means aggregate program design should explicitly account for the composition of who sits at the margin under any given access policy. The processing-time concern is underappreciated: if 125 extra days of appeals processing generates skill depreciation that suppresses labor supply for strictly-assigned denied applicants, then the conventional "denied applicants as control group" strategy (Bound 1989) provides neither an upper bound (Parsons 1991) nor a clean estimate of the benefit receipt effect — it conflates benefit receipt with processing time. See SSA Sequential Determination Process for the appeals timeline context.