Autor Maestas Mullen and Strand 2015 — Does Delay Cause Decay The Effect of Administrative Decision Time on the Labor Force Participation and Earnings of Disability Applicants

Summary

Using a universe of $1{,}039{,}221$ 2005 Social Security Disability Insurance (SSDI) initial determinations (Disability Insurance Operational Data Store (DIODS) database), the authors identify two distinct channels through which the disability application process affects labor force participation: the benefit receipt effect and the processing delay effect. Separate jackknife examiner instruments for processing time (EXTIME) and allowance propensity (EXALLOW) — which are nearly orthogonal ( $\rho = 0.016$ ) — allow the paper to jointly estimate both effects in a single system. Each additional month of administrative processing time reduces employment by $0.44$ – $0.52$ percentage points (pp) and annual earnings by $\$133$ three years post-decision, with effects persisting 6+ years; a falsification test using the 5-month statutory waiting period confirms the mechanism is labor force atrophy rather than examiner quality. The corrected SSDI receipt estimate ( $-48$ pp employment at year 3) is roughly $80\%$ larger than the Maestas, Mullen, and Strand (MMS) (2013) single instrumental-variable (IV) estimate of $-27$ pp, because the EXALLOW-only IV conflates the receipt effect with the opposite-signed delay-channel effect for denied applicants.

Key Claims

Delay-decay effect: +1 month processing time $\to$ employment $-0.44$ pp (year 2), $-0.52$ pp (year 3); substantial gainful activity (SGA) earnings $-\$133$ per year (year 3); effects persist through year 6+; SGA employment $-0.31$ pp (year 3)
Mechanism confirmed by falsification: Delay effects are zero during the statutory 5-month waiting period (months 1–5 of processing); effects emerge only after month 5. Applicants cannot earn above SGA during the waiting period regardless of labor force attachment — the atrophy occurs only once LFP choices open up, confirming the channel is human capital decay from non-participation, not an examiner quality artifact
Corrected SSDI receipt effect: SSDI receipt $\to$ $-48$ pp employment (year 3), $-25$ pp (year 6), using joint two-IV system; vs. $-27$ pp in MMS (2013) using EXALLOW alone
Bias formula: $\gamma^{\text{MMS}} = (\gamma \pi_1^d + \delta \pi_1^t)/\pi_1^d$ . EXALLOW is correlated with both benefit receipt (first stage $\pi_1^d$ ) and processing time ( $\pi_1^t < 0$ — lenient examiners decide faster). Since $\delta < 0$ (more processing months reduce employment) and $\pi_1^t < 0$ , the term $\delta \pi_1^t / \pi_1^d$ is positive, attenuating the single-IV estimate of $\gamma$ toward zero. The EXALLOW-only reduced form reflects $\gamma + \delta(\pi_1^t/\pi_1^d)$ , not $\gamma$ alone
Combined program effect: Delays reduce denied applicants' labor force participation (LFP) by $2.4$ pp ( $5.8\%$ ) and allowed applicants' LFP by $3.1$ pp ( $36.5\%$ ). Total SSDI program effect (receipt + delay) is $105$ – $150\%$ larger than prior literature estimated
Random assignment validation: Within–Disability Determination Services (DDS)–office examiner assignment is conditionally random (conditional on body system, terminal illness (TERI) flag, top 20 diagnoses). Applicant characteristics are orthogonal to EXTIME and EXALLOW conditional on office-by-period fixed effects; EXTIME and EXALLOW are nearly uncorrelated with each other ( $\rho = 0.016$ )
Sample: $N = 1{,}039{,}221$ ; ages 18–64; 2005 initial Social Security Administration (SSA) decisions (DIODS); avg processing time $14.1$ months; $33.1\%$ initially allowed, $67.7\%$ finally allowed including appeals

Concepts Introduced or Extended

Entities Mentioned

Quotes

"The upshot is that the SSDI program, combining both the benefit receipt and application processing time effects, reduces long-run labor supply by 105 to 150 percent more than estimated in the prior literature."

"The mechanism that we propose is that when applicants exit the labor force to await a decision, they experience a depreciation of their human capital — their skills atrophy and their connections to employers weaken."

My Take

This paper makes two contributions of unequal difficulty. The delay-decay channel is new and compelling; the 5-month waiting period falsification is elegant and persuasive. The corrected receipt effect ( $-48$ pp) is the more consequential result — it substantially revises the MMS (2013) benchmark — but it rests critically on the near-orthogonality of EXTIME and EXALLOW ( $\rho = 0.016$ ). If examiners who process slowly also differ systematically in who they allow (even slightly), the identification fails. The paper checks this carefully and the exclusion restriction appears to hold, but the result warrants replication on other cohorts. The DIODS 2005 cohort is a single year; the 1997–2002 MMS cohort produced $-27$ pp; it is not fully resolved how much of the gap reflects identification improvement versus cohort composition differences.