Summary
This five-page AEA Papers and Proceedings article estimates the causal effect of the Social Security Disability Insurance (SSDI) application process — independent of receiving benefits — on the subsequent employment of denied applicants. Using Health and Retirement Study (HRS) data merged with Social Security Administration (SSA) administrative records (11 waves, 1992–2012), and an instrumental variables (IV) design exploiting Full Retirement Age (FRA) variation across birth cohorts and state-level Disability Determination Services (DDS) allowance rates, Khan finds that denied applicants aged 50–58 suffer a 36 percentage point (pp) reduction in employment two to three years after filing. The finding challenges the Bound (1989) "denied applicants as control group" tradition by showing that denied applicants are themselves "treated" by the process — not a clean counterfactual for accepted applicants.
Key Claims
- Application-process employment effect (causal): The SSDI application decision itself causes a 36 pp reduction in employment of denied applicants aged 50–58, measured 2–3 years post-application (IV/two-stage least squares [2SLS]; standard error [SE] = 0.20). Ordinary least squares (OLS) upper bound: −49 pp.
- Three channels (Bound 1989; Parsons 1991): (i) Denied applicants voluntarily exit the labor force while planning to reapply; (ii) appellants stay out of work during appeals to strengthen their case for adjudication; (iii) human capital deteriorates during extended absence, making re-entry harder once the process ends.
- Control group innovation: Rather than using all non-applicants (very different health and hassle-cost profiles), Khan restricts controls to future applicants — individuals who did not apply in their 50s but eventually filed at age 60+. These have comparable health trajectories (likely experiencing deterioration in their 50s) and intermediate "hassle cost" — high enough to defer application, but not so high that application is never worthwhile. 322 denied applicants vs. 347 controls; 1,231 observations total.
- IV instruments: (1) FRA variation across birth cohorts from the 1983 Social Security Amendments: higher FRA makes SSDI relatively more generous than Old-Age and Survivors Insurance (OASI) at 62, raising application incentives. Cohorts with FRA ≥ 66 are 17 pp more likely to be denied applicants vs. FRA = 65 cohorts (significant at 1%). (2) State-level DDS allowance rate. First-stage F = 9.84 (above 9.08 critical value for max 10% bias); overidentification test p = 0.57.
- Challenge to Bound (1989): If the application process itself depresses denied applicants' employment, the accepted–denied comparison understates the causal employment effect of benefit receipt — denied applicants are also "treated" by process rather than benefit.
- Policy stakes: Denial rate rose from 45% (2000) to 72% (2013); 1.8 million denied in 2013 alone. The 36 pp process-induced employment loss on a growing denied population represents a large and growing aggregate welfare cost. Recommendation: shorten the determination process and reduce work disincentives during the application and appeal waiting period.
Concepts Introduced or Extended
- Conditional DI Applicants — complicates the denied-applicants-as-control-group assumption; the application process itself reduces denied applicants' employment, contaminating accepted–denied comparisons
- DI Denied Population — new mechanism: the application process imposes employment costs on the denied even absent benefit receipt
Entities Mentioned
Quotes
"The loss of potential employment of the denied SSDI applicants is a welfare loss to the society that results due to their decision to go through the process of application."
"More people than ever are applying for disability benefits, removing themselves from the labor market for months or even years, and then being forced to re-enter when they are denied enrollment."
My Take
A compact and methodologically precise contribution. The 36 pp causal estimate is striking from a sample of only 1,231 observations; the IV is well-powered by the FRA discontinuity, but the standard error (0.20) is large, and the paper establishes the sign and order of magnitude rather than a precise effect size.
The control-group design is the most innovative aspect: future applicants (those who filed in their 60s) are a more defensible comparison group than never-applicants, avoiding the critique that the two groups differ systematically in health and preferences. That said, the design cannot fully rule out differential health deterioration dynamics in the 50s between the treatment and control groups.
The application-process welfare loss documented here complements adjacent findings — Autor, Maestas, Mullen, and Strand (2015) decay from processing delay, French and Song (2014) earnings suppression during appeals — but isolates the application-decision channel specifically for denied applicants as a standalone causal estimate. The aggregate welfare cost is large: with 1.8 million annual denials and a 36 pp employment effect, the total employment loss to society from the process itself is economically significant and growing.