Using quasi-random assignment of disability insurance (DI) appeal cases to Administrative Law Judges (ALJs) as an instrument, French and Song estimate the causal effect of DI benefit receipt on labor supply. The judge allowance differential — each judge's leave-one-out mean allowance rate, demeaned by hearing office and day — predicts allowance with a first-stage coefficient of 0.764 (t=101). The sample covers 1.78 million ALJ hearings from 1990–1999 with earnings tracked through 2009. DI receipt reduces labor force participation (LFP) by 26 percentage points (pp) and annual earnings by ~$4,059 three years after ALJ assignment, rising to $4,915 at five years. A key secondary finding: over 60% of ALJ-denied applicants ultimately receive benefits within ten years, showing that denial is rarely a terminal outcome.
| Group | LFP effect (3 yr, IV) |
|---|---|
| Age 60–64 | −6.9 pp |
| College graduates | −17.9 pp |
| Mental disorders | −20.2 pp |
| Age ≤45 | −29.0 pp |
| Musculoskeletal | −28.5 pp |
| Injuries | −36.7 pp |
"Over 60 percent of those denied benefits by an administrative law judge are subsequently allowed benefits within ten years, showing that most applicants apply, reapply, and appeal until they get benefits."
"OLS and instrumental variables estimates are similar."
This paper is the ALJ-stage complement to Maestas, Mullen, and Strand (2013), which uses Disability Determination Services (DDS) examiner assignment at the initial determination stage. Both find large causal work-disincentive effects (~26–28 pp LFP reduction), and together they establish that the effect is consistent across two different stages of the adjudicative process and two different complier populations.
The OLS≈IV finding deserves emphasis: it implies that at the ALJ appeal stage, simple comparisons of allowed vs. denied applicants yield approximately unbiased estimates of the treatment effect. The selection biases (sicker people more likely to be allowed; people planning to work anyway more likely to be denied) roughly cancel at this stage. This is a nontrivial result that strengthens the credibility of all observational studies on this population.
The "60% eventually allowed" finding is the paper's most significant policy contribution. It directly undermines the standard framing of DI denial as a fiscal cost-control mechanism. If most denied applicants eventually receive benefits after years of appeals — while holding their earnings below SGA during the process — then denial delays benefit receipt and imposes costs (earnings suppression, legal fees, health deterioration during waiting) without permanently excluding many applicants. The Autor, Maestas, Mullen, and Strand (2011) "decay effect" ($133 lower earnings per additional month of processing time) compounds this harm.
The subgroup heterogeneity is the clearest guide for policy reform: the near-zero effect for ages 60–64 reflects that they would largely exit the labor force at retirement anyway; the near-zero effect for college graduates and those with mental illness reflects genuine inability to work. The large effect for younger workers and those with injuries suggests this is the subpopulation for whom work-support interventions — rather than benefit provision — would be most effective. See DI Denied Population and Conditional DI Applicants.