The paper develops a bounds-based decomposition of Social Security disability insurance (DI) enrollment dynamics across two distinct episodes — the historical rise (1988–2010) and the post-peak decline (2010–2019). Unlike prior sequential counterfactual frameworks (Liebman 2015; Pattison and Waldron 2013), it accounts for interaction terms between five enrollment margins (population composition, program eligibility, application, award, and exit), preventing double-counting in a sequential decision problem. On the decline, it finds that reduced applications — not tighter eligibility — explains roughly 71% of the enrollment fall, with labor demand recovery as the dominant causal mechanism (two-stage least squares (2SLS) estimate: 1 percentage-point (pp) rise in employment/population → −0.368 pp annual application rate). Administrative Law Judge (ALJ) reform and field office closures account for at most a few percent each.
Enrollment accounting identity: , where = enrollees, = entry rate, = exit rate, = population in group at time . Decomposing growth requires attributing joint changes across all five margins simultaneously; interaction terms cannot be assigned to any single margin without bias.
Bounds approach: Because the margins interact (e.g., a larger eligible population simultaneously changes application and award margins), the paper uses partial-identification bounds rather than point decompositions. This bounds the contribution of each margin to the enrollment change without requiring arbitrary interaction attribution rules.
The rise (1988–2010): The application margin is the dominant factor explaining enrollment growth. Demographic forces (population growth and aging into disability-prone years) play a secondary role. Award rates and exit rates contribute modestly in the expected direction but are not the primary drivers.
The decline (2010–2019) — naive decomposition: Award rates appear to account for ~42% and applications ~58% of the enrollment decline. This overstates the role of award rates because the applicant pool became compositionally less severe over time — as labor market recovery drew healthier conditional applicants back to work, the remaining applicant pool had lower observable severity, mechanically reducing award rates even without any policy tightening.
The decline — composition-adjusted decomposition (Eq. 13): After adjusting for changes in applicant composition, applications account for ~71% and award rates ~29% of the decline. The award rate decline is substantially a compositional artifact, not a genuine eligibility tightening.
Labor demand as causal mechanism (shift-share instrumental variable [IV]): Using a Bartik-style shift-share instrument (Autor-Duggan design) exploiting variation in local industry employment mix, the 2SLS estimate is: 1 pp increase in employment-to-population ratio → −0.368 pp reduction in annual application rate. This labor demand channel explains approximately 73% of the application decline 2010–2019.
ALJ reform (2017–2019): Award rates at the ALJ stage fell sharply during the Trump administration's changes to ALJ adjudication (stricter standards, increased documentation requirements). The paper tests whether this award-rate decline deterred applications — finding an elasticity of applications to ALJ award rates approximately equal to zero. Applications did not respond to the ALJ award rate fall. This falsifies the hypothesis that applicants at the margin optimize over expected ALJ award probabilities; the ALJ reform accounts for at most ~0–3% of the total enrollment decline.
Field office closures: Social Security Administration (SSA) field office closures 2010–2019 account for approximately 3–6% of the enrollment decline — small relative to the labor demand channel.
Population health ruled out: Changes in underlying disability prevalence or population health cannot account for the magnitude or timing of the enrollment decline. Disability rates in the general working-age population did not improve sufficiently between 2010 and 2019 to explain the observed decline.
Demographic heterogeneity: The enrollment decline was concentrated among workers aged 50–64, workers with lower education, and conditions in the musculoskeletal category — precisely the demographic profile of conditional applicants who prefer work when employment is available and apply when it is not.
"After adjusting for the composition of applicants, reduced applications account for approximately 71 percent of the enrollment decline, with award rates accounting for the remaining 29 percent."
"A one percentage point increase in the employment-to-population ratio reduces the annual DI application rate by 0.368 percentage points."
"We find essentially no response of applications to the decline in ALJ award rates during the reform period — suggesting that applicants do not condition their application decisions on expected award rates at the ALJ stage."
This paper is the most rigorous decomposition of the Social Security Disability Insurance (SSDI) enrollment arc to date. The interaction-aware bounds framework is a genuine methodological advance over both Liebman (2015) and Pattison-Waldron (2013), which apply sequential counterfactuals that implicitly attribute interaction terms to specific margins. The composition-adjustment for the award rate (the Eq. 13 correction) is a particularly important contribution: it prevents policy analysts from misreading a mechanical award-rate decline as evidence of successful eligibility tightening when the real cause is that healthier conditional applicants left the pool first. The ALJ-reform null result is important for program integrity debates — it suggests that tightening adjudication standards does not deter applications, only recipients; the welfare consequences depend on whether the additional denied cases were truly ineligible or merely screened out inappropriately. The paper's findings on the decline complement Deshpande and Li (2019) on the rise: both show that application behavior — not eligibility standards — is the primary margin through which enrollment is determined.