Conditional Disability Insurance (DI) applicants are individuals who have a qualifying health impairment and could potentially receive Social Security disability benefits, but prefer to work when employed. They apply for DI only after losing their jobs — making their applications conditional on labor market conditions. This contrasts with inframarginal applicants, who experience sudden, severe disabilities and apply regardless of the economic environment. The typology was coined in Autor and Duggan (2002, CRR WP), formalized in the published version (2003, QJE), and is empirically confirmed at the business cycle frequency by Lindner, Burdick, and Meseguer (2017).
Since the mid-1980s, three forces have expanded the conditional applicant pool:
This structural expansion is documented by Liebman (2015) via the musculoskeletal/mental condition shift and the post-1985 incidence bounce-back. See DI Growth Decomposition.
At the business-cycle frequency, recessions activate the existing conditional applicant stock without necessarily expanding it. The mechanism:
Using the universe of all DI applications 1991–2008 (22.7 million, SSA Disability Research File), Lindner et al. find:
A second empirical fingerprint comes from the onset-to-filing gap (Maestas, Mullen, and Strand 2015): using Electronic Disability Collect System (EDCS) alleged onset dates, a 1 pp rise in unemployment is associated with +0.512 months (≈2 weeks) in the median time between alleged disability onset and filing. The gap is concentrated among denied applicants (+0.589 months/pp) and almost nonexistent for medical listing allowances (+0.073 months = 2.2 days). During the Great Recession, the median gap grew by 2.5 months — recession-era applicants had been struggling with their impairments for longer before applying. Crucially, Disability Determination Services (DDS) examiners adjust alleged onset dates to established onset dates, removing nearly all of this cyclicality (established onset increases by only 0.056 months/pp). SSA is not granting extra back-pay to recession-era applicants.
The conditional applicant model treats all non-inframarginal applicants as having had employment alternatives that were removed by labor market conditions. Contreary, Honeycutt, Stegman Bailey, and Mastrianni (2017) document that even within the non-inframarginal pool, pre-application employment histories are highly heterogeneous — a distinction with direct implications for early-intervention policy.
Using Survey of Income and Program Participation (SIPP) panels (1996, 2001, 2004) linked to SSA administrative records for applicants ages 25–55, they characterize employment in the 6–24 months before application:
| Employment pattern (6-month window) | Share | Allowance rate |
|---|---|---|
| Consistently employed | 28% | ~47–50% |
| Ceased employment | 25% | ~47–50% (indistinguishable from above) |
| Intermittent employment | 14% | ~37–40% (−10 pp vs. consistent) |
| No employment | 33% | ~32–37% |
This yields a two-type summary:
Key quantitative findings:
Policy implication: Employer-focused early-intervention and worker-retention proposals (which target recently displaced workers before they apply) miss approximately half the DI applicant pool. Type 2 applicants are reachable primarily through the public programs — Medicaid, SNAP — they already participate in. Any comprehensive DI diversion strategy requires two distinct intervention points: the workplace for Type 1 and the public program network for Type 2.
Complementing the Contreary et al. employment history typology, Lindner (2013) classifies non-employment spells of DI applicants by their beginning event — the precipitating circumstances of job separation — and documents sharp heterogeneity in application timing, job search, and Unemployment Insurance (UI) receipt across types. Data: SIPP 1990–2004 matched to SSA administrative records; non-employment subsample n = 1,246 (SIPP 1996–2004: n = 705).
Pre-application employment pattern: Employment among future DI applicants runs at ~80% thirty months before application, falls to ~60% at ten months out, then plunges to ~30% at the application month. Job loss explains nearly all of the income decline preceding application (decomposition shows employment changes track income changes almost perfectly). Over 30% of applicants are out of work 6+ months before applying — confirming that the pre-application non-employment spell, not just the health shock, matters for the application decision.
Beginning-event frequencies (SIPP 1996–2004):
Spell characteristics by beginning event (reference = job loss; negative binomial for counts, logit for binary; controls for demographics, industry, year):
| Beginning event | Spell duration | Waiting time | UI receipt | Job search |
|---|---|---|---|---|
| Illness/injury | ~4.8 months (−57%) | ~6.7 months (−23%) | 11% | Very low |
| Work limitation | ~6.2 months (−32%) | ~7.5 months (−14%) | 32% | Moderate |
| Job loss (ref) | ~9.2 months | ~8.7 months | 71% | High |
| Quit | ~7.9 months | ~9.0 months | 47% | Low |
Application success rate is numerically higher for illness/injury (69% vs. 60% for job loss) but not statistically significant, suggesting the acute/chronic distinction is more salient for timing and process characteristics than for ultimate award probability.
Policy mapping onto the two-type framework: Illness/injury applicants (~49%) apply quickly, rarely job search, and are unlikely to respond to re-employment incentives — they are the closest empirical analog to inframarginal applicants. The remaining ~51% (work limitation, job loss, quit) exhibit extended non-employment with active job search and UI use before applying — this group is the target for early intervention. This typology converges with Contreary et al.'s Type 1/Type 2 classification from a different angle: Contreary et al. use retrospective employment history; Lindner (2013) uses contemporaneous beginning events. Together they triangulate the same ~50/50 split. See Lindner 2013 — From Working to Applying Employment Transitions of Applicants for Disability Insurance in the United States.
Administrative earnings data reveal a systematic difference in the shape of the pre-application earnings decline between allowed and rejected applicants — a distinction that maps directly onto the conditional vs. inframarginal typology:
This pre-application trajectory difference is a non-parametric fingerprint for the two applicant types using administrative data alone, without requiring structural modeling. The gradual multi-year dip for rejected applicants also directly explains why these workers are hard to distinguish from conditional applicants until the moment of application: their impairments have been worsening for years before their employment alternative finally disappears. See Von Wachter Song and Manchester 2011 — Trends in Employment and Earnings of Allowed and Rejected Applicants to the Social Security Disability Insurance Program.
Denied applicants who continue appealing face a strong incentive to suppress their earnings below the Substantial Gainful Activity (SGA) threshold (1,040 in 2013) to preserve eligibility during the ongoing adjudication. This is rational forward-looking behavior: an applicant who earns above SGA during appeals is presumed capable of substantial gainful activity and will be denied. French and Song document that this suppression is widespread — 75% of Administrative Law Judge (ALJ)-denied applicants are either allowed or still actively appealing 3 years after assignment — and that it attenuates the measured work-disincentive at short horizons (the effect grows from −4,915 at 5 years as appeals resolve). This is distinct from, but reinforces, the Autor/Maestas/Mullen/Strand "decay effect": processing delay reduces employment via both the earnings-suppression channel (strategic) and the attachment-decay channel (structural).
The earnings suppression mechanism (French and Song 2014) documents strategic behavior during appeals. Khan (2018) estimates the total employment cost of the application decision itself — combining all channels — for applicants who are ultimately denied.
Research design: Health and Retirement Study (HRS) 1992–2012 (11 waves), merged with SSA administrative geographic data. Treatment group: 322 denied applicants aged 50–58. Control group: 347 future applicants — individuals who did not apply in their 50s but eventually filed at age 60+. This control group innovation addresses the Bound (1989) limitation: future applicants have comparable health deterioration trajectories and intermediate "hassle cost," making them closer comparators than never-applicants who may simply be healthier or more averse to the program. Instrumental variables (IV): (1) Full Retirement Age (FRA) variation across birth cohorts from the 1983 Social Security Amendments (cohorts with FRA ≥ 66 are 17 pp more likely to be denied applicants vs. FRA = 65; significant at 1%); (2) state-level DDS allowance rate. First-stage F = 9.84; overidentification test p = 0.57.
Results:
Three channels embedded in the estimate: (i) voluntary labor force exit while planning reapplication; (ii) voluntary exit during appeal to strengthen the case; (iii) human capital deterioration from extended absence that makes re-entry difficult once the process ends.
Methodological implication for the accepted–denied comparison: The Bound (1989) tradition uses denied applicants' labor supply as an upper bound for what accepted applicants would have earned absent benefits. If the application process itself reduces denied applicants' employment by 36 pp, this upper bound is contaminated — denied applicants are also "treated" by process, not just by benefit receipt. The accepted–denied comparison understates the causal employment effect of benefit receipt by an unknown but potentially large amount.
Policy context: Denial rate rose from 45% (2000) to 72% (2013); 1.8 million denied in 2013. The process-induced employment loss is a growing aggregate welfare cost. Khan recommends shortening the determination process and reducing work disincentives during the application waiting period. See Khan 2018 — Disability Insurance Application Decision.
Conditional applicants have higher work capacity — which implies better post-application employment for denied applicants during recessions. But adverse labor market conditions simultaneously suppress those outcomes. Empirically, the labor market effect dominates slightly:
The composition effect (conditional applicants → higher capacity → better outcomes) is more than offset by the recession's direct suppression of labor demand.
A naive reading: more conditional applicants during recessions → DI draws more people away from work who could otherwise be employed → higher disincentive effect → program is more wasteful during recessions.
Lindner et al. show this is wrong. The disincentive effect — estimated as counterfactual earnings of accepted applicants had they not enrolled — is slightly negatively related to unemployment. Recessions suppress labor demand for the entire low-skill workforce, including conditional applicants. The foregone earnings from DI participation are lower during recessions because the employment alternative is less available and less remunerative. The program is a less distortionary insurance mechanism precisely when it is most heavily used.
The policy implication: equating "higher share of conditional applicants during recessions" with "higher disincentive effect" confounds composition with labor market conditions. They must be identified separately.
Many recession-era applicants are quickly rejected at Steps 2 and 4 (higher work capacity → early denial) and then enter the lengthy appeals process before eventually being accepted via re-application. Acceptance rates:
This roundabout path may further erode employment prospects — the "decay effect" (Autor, Maestas, Mullen, and Strand 2015): a one-month increase in processing time reduces employment by 0.47 pp and earnings by $133 three years after decision. Short-term financial support plus re-employment services at the point of initial rejection would be a more efficient intervention for this population than routing them through the multi-year appeals process.
The composition shift toward higher-work-capacity applicants during recessions is sometimes cited as evidence that DI is accepting people who could work. The Lindner et al. framework reframes this: these applicants are concentrated precisely among those initially rejected, and their post-application labor market outcomes are worse during recessions not because they lack capacity but because labor markets are depressed. The program is responding to genuine economic distress, not to looser screening.
Three papers form the definitive arc on identifying the mechanism behind cyclical activation of the conditional applicant stock:
Cutler, Meara, and Richards-Shubik (2012) use Health and Retirement Study (HRS) micro data (men 52–64, 1992–2010) and test both leading mechanisms. 1 pp unemployment rate (UR) → +4.8/1000 application probability. Controls for health severity (lung disease, cancer, psychiatric conditions, functional limitations) do not reduce this coefficient — health shocks are rejected as the mechanism. Interactions with low education, recession-affected industries, and near-retirement status also fail to attenuate the UR effect — the standard opportunity-cost proxies are rejected too. The paper clears the field without resolving the mechanism, and notes extended Unemployment Insurance (UI) benefits as a possible unidentified explanation.
Cutler et al. (2015) (ages 55–61, HRS 2000–2012) narrow the population: the entire ~30% Great Recession (GR) application rise is concentrated in workers with functional limitations (FL). Application rate among those with 1+ FL: ~5–6.5%; without FL: ~0.3–0.5%. FL prevalence is recession-invariant, so the conditional application rate rises within the FL group. A cohort analysis confirms that GR-onset FL workers applied faster (9% within one HRS wave) than pre-recession FL cohorts, but their labor force participation, earnings, and full-time work were statistically identical — the recession accelerated DI timing without generating new disability or earlier work exit. The conditional applicant model is confirmed at the FL level; the mechanism (health shock vs. opportunity cost) remains open.
Carey, Miller, and Molitor (2025) provide the definitive mechanism test using Medicare data and the age-50/55 Vocational Grid discontinuity as an instrument for pure entry-cost variation. Recession entrants are healthier (lower spending, lower mortality) than typical recipients. The marginal spending functions at mean and high unemployment are statistically indistinguishable: health shocks explain only 1.5% of DI cyclicality; entry-cost reduction accounts for 92.8–98.5%. The conditional applicant model is confirmed at the mechanistic level — recession activation is overwhelmingly an opportunity-cost effect, not new disability. See DI Countercyclicality.
A key prediction of the conditional applicant model is that an economically-driven expansion of the applicant pool should not be accompanied by health deterioration — marginal applicants enter because economic conditions push them in, not because they became sicker. Rutledge, Wu, Guan, and Trenkamp (2014) confirm this directly: measuring Social Security Disability Insurance (SSDI)/Supplemental Security Income (SSI) applicants' health 1–3 years before application using SIPP-SSA linked data (1990–2013), they find flat or improving pre-application health across most measures. Work-preventing conditions declined; activities of daily living / instrumental activities of daily living (ADL/IADL) limitations and self-reported poor health were unchanged. The pool expanded without becoming healthier — exactly what the conditional applicant model predicts if growth is demand-driven. See DI Growth Decomposition.
The conditional applicant mechanism requires a labor demand shock to activate the latent pool of health-impaired workers who prefer work when employed. Autor, Dorn, and Hanson (2019) provide the causal identification of the specific shock responsible for a large share of manufacturing-sector displacement over 1990–2014: rising Chinese import competition.
Using gender-specific Bartik instruments, ADH show that a one-unit trade shock to male-intensive manufacturing:
The idle male population ADH document is the precursor to DI applications: health-impaired men who have lost their manufacturing jobs and cannot find alternative employment are precisely the conditional applicant stock that Autor and Duggan (2003) characterized. ADH do not track DI applications directly, but the upstream mechanism is the same China-shock displacement that DI Growth Decomposition literature has linked to incidence growth.
This connection is corroborated by Michaud et al. (2017): the demographic group with the strongest trade-shock exposure — workers aged 55–59 without HS degrees in manufacturing/production — is also the group most responsible for vocational-stage DI awards. The China trade shock thus provides causal upstream identification for the conditional applicant flow. See China Trade Shock.
The 5-month non-employment requirement before Social Security Disability Insurance (SSDI) filing means workers in the application pipeline are structurally outside employment counts even before they are awarded — creating what Michaud and Wiczer (2018) call the SSDI non-employment shadow.
Scale: At any point in time, 1.6–2.5% of the working-age population is in the SSDI application pipeline. Among non-employed workers specifically, the fraction applying to SSDI grew from 25% (1985) to 66% (2010). This means that by 2010, the majority of non-employed prime-age workers were in some stage of SSDI application.
Measurement implication: Empirical studies that estimate SSDI's employment effect by comparing awarded vs. denied applicants miss this pipeline population. Omitting pipeline applicants understates the measured employment effect of SSDI by 30–40%. The true labor market footprint of SSDI is substantially larger than award-based comparisons suggest.
Composition within the shadow: The non-employment shadow population is not equivalent to the conditional applicant stock. It includes: (a) workers who have already determined to apply and are serving the non-employment waiting period; (b) workers who have filed but are awaiting initial determination; (c) denied applicants in the appeals pipeline who must suppress earnings to preserve eligibility (see Earnings Suppression section below). All three groups are structurally absent from employment statistics during their pipeline spell.
Structural vs. cyclical shadow: In the structural model, the shadow is persistent and growing — driven by secular wage trends and aging. At the cyclical frequency, recessions increase the flow into the shadow (more job loss triggers more application entry) but the 5-month non-employment threshold smooths the surge. This is consistent with Maestas, Mullen, and Strand (2015) finding that cyclical application elasticities are moderate — the non-employment waiting period acts as a buffer between job loss and formal application.
See DI Growth Decomposition for the Michaud-Wiczer Shapley-Owen decomposition confirming that business cycles have near-zero impact on awards despite large application effects.
The conditional applicant model's core structural prediction — that permanent, not transitory, earnings losses drive DI participation — receives its earliest rigorous causal test from Black, Daniel, and Sanders (2002). Using the coal boom (1970s, Organization of the Petroleum Exporting Countries (OPEC)-driven) and bust (1980s) in Kentucky, Ohio, Pennsylvania, and West Virginia, they instrument local earnings with the interaction of the log annual coal price (set on world markets, exogenous to any county) with the log of county coal reserves (geological constants). This variation is orthogonal to transitory income fluctuations and captures only the permanent component of local earnings prospects.
The Ordinary Least Squares (OLS)≈0 / Instrumental Variables (IV) strong divergence is the key result. OLS estimates of the DI payments elasticity are near zero and insignificant — annual earnings contain large transitory fluctuations that dominate the signal. Two-Stage Least Squares (2SLS) estimates are −0.35 to −0.41 (DI) and −0.40 to −0.71 (Supplemental Security Income (SSI)), all statistically significant (first-stage F ≈ 27). Steel industry replication (6 states, 1982–87) confirms generalizability: DI = −0.361, SSI = −1.049.
UI placebo test confirms the mechanism. The same IV has a near-zero effect on UI payments per capita, while OLS for UI is positive and significant. The pattern is structurally diagnostic: UI insures transitory spells (OLS strong, IV≈0); DI and SSI insure permanent earnings losses (OLS≈0, IV strong). The programs are differentiated by the time horizon of the shock they insure, and the data confirm this cleanly.
SSI > DI elasticity reflects the means-test dual channel: SSI eligibility depends on both own earnings and family income, creating two pathways for permanent earnings changes to affect participation. DI only operates through the own-earnings replacement-rate channel. The differential quantifies the additional responsiveness from the family-income mechanism.
The behavioral logic follows from DI's quasi-permanence (~1.6% annual exit rate, ~2-year time-to-benefit): workers who optimize over lifetime income condition their application decision on expected permanent earnings, not current-year fluctuations. Black et al. provide the earliest econometric validation of this behavioral assumption — the foundation for Autor and Duggan's (2003) replacement-rate framework and Deshpande et al.'s (2025) shift-share confirmation of the same mechanism in the post-2010 decline. See Black Daniel and Sanders 2002 — The Impact of Economic Conditions on Participation in Disability Programs Evidence from the Coal Boom and Bust.
The conditional applicant mechanism predicts that as labor demand recovers, the latent pool of health-impaired workers will return to employment rather than apply — generating a symmetric decline in DI applications that mirrors the recession-era surge. Deshpande, Kellogg, Mogstad, and Tseng (2025) provide the first causal identification of this mechanism for the post-2010 decline using a Bartik shift-share IV (Autor-Duggan design) that exploits variation in local industry employment mix.
Key causal estimate:
Implications for the conditional applicant model:
The ALJ-reform test: If conditional applicants optimize over expected award probabilities (not just employment alternatives), then a fall in ALJ award rates should also deter applications. Deshpande et al. test this with the 2017–2019 ALJ reform, which sharply reduced ALJ award rates. Applications did not respond — the elasticity of applications to ALJ award rates is approximately zero. This is consistent with a model where conditional applicants condition their application decision on local labor market conditions and initial-stage award expectations, but do not track ALJ award rates — perhaps because the ALJ stage is too far removed from initial filing and too uncertain to influence entry-margin behavior.
The Autor-Duggan (2003) model implies that a nontrivial fraction of recession-era SSDI applicants are displaced workers who exhaust UI benefits and then apply for SSDI — the "UI-before-SSDI" type. Mueller et al. (2016) directly test this using matched Current Population Survey Annual Social and Economic Supplement (CPS ASEC) panels (2005–13 surveys) that link pre-award labor force attachment to new SSDI receipt. Their finding imposes a tight empirical upper bound on this sub-type:
This means the Autor-Duggan "UI-before-SSDI" type comprises at most 28% of SSDI awardees — and even within that 28%, most do not use UI as a bridge. The model's predicted elasticity of SSDI applications with respect to UI exhaustion (0.5) must be scaled down by roughly 4×, which is consistent with the near-zero empirical elasticities Mueller et al. find across all three identification strategies (time-series, state panel FE, event studies around UI extensions). The panel estimate rules out contemporaneous elasticities larger than 0.005.
Implication for the cyclical mechanism: Since UI exhaustion does not drive the recession-era SSDI surge, the conditional applicant pool's activation during recessions must operate primarily through channels other than the UI exhaustion pathway — most likely via employer accommodation failures, relative generosity increases (wages fall in recessions; SSDI benefits do not), or direct labor force exit without a UI spell. See DI Growth Decomposition.
Obesity is a structural health-demand driver that creates latent conditional DI applicants: individuals whose obesity-induced musculoskeletal or metabolic limitations worsen progressively but who remain employed until a job loss or health shock removes their employment alternative. Burkhauser and Cawley (2004) provide causal evidence using 2SLS (instrument: biological relative's weight) on Panel Study of Income Dynamics (PSID) and National Longitudinal Survey of Youth 1979 (NLSY79) data for ages 25–44: obesity raises disability income receipt by 5–9 pp, with OLS underestimating this by 5–15× due to measurement error and endogeneity.
This finding is consistent with the conditional applicant mechanism. Because obesity-related impairments (musculoskeletal conditions, diabetes, cardiovascular risk factors) typically develop slowly and are partially managed for years before they prevent sustained work, they create exactly the profile of a conditional applicant: a health-impaired worker who prefers employment but whose physical capacity erodes over time, making DI application rational after job loss or when health deteriorates sufficiently. The parallel doubling of obesity rates and DI rolls since the early 1980s is consistent with obesity enlarging the latent conditional applicant stock over this period. See DI Growth Decomposition and Morbidity-Mortality Distinction.
A distinct mechanism activating the conditional applicant pool is workplace injury. O'Leary et al. (2012) — using matched New Mexico Workers' Compensation Authority (WCA) records (1994–2000) linked to SSA administrative data — document that a lost-time workplace injury roughly doubles the 10-year DI receipt probability (6% medical-only → 12% lost-time). The key result for the conditional applicant framework: the excess DI hazard is not concentrated in workers classified as permanently and totally disabled by Workers' Compensation (WC) (only 0.5% of lost-time cases). Even workers with temporary disability of less than 8 weeks show substantially elevated DI risk compared to medical-only controls.
This pattern is exactly what the conditional applicant model predicts:
The "10-year aging effect" (a lost-time injury ≈ 10 years of additional disability risk) is consistent with injury-induced conditional applicants aging into DI enrollment at the same rate as an older uninjured cohort — suggesting that injury accelerates the transition from "working despite impairment" to "unable to maintain employment despite attempting to."
See Workers' Compensation and DI for full data on the WC-DI pathway, covariate gradients, and fiscal implications.
Mental illness, particularly depression, is a distinct conditional applicant pathway that prior economics literature missed by using aggregate mortality as the health proxy for DI trends. Conti, Berndt, and Frank (2009) use HRS panel data (birth cohort 1931–1941, waves 2–5: 1994–2000) to estimate both direct and indirect effects of depression on DI/SSI applications and early retirement.
Direct effect: Depression alone raises DI/SSI application probability by approximately 16% for men — comparable in magnitude to incident physical illness (~25%). Both Generalized Estimating Equations (GEE) estimation strategies confirm this: the effect holds for women at a similar magnitude. This means the standard physical illness framing in DI research has overstated that pathway's standalone causal role, because the mental health channel was excluded.
Depression × physical illness (indirect): Depression interacted with physical illness raises DI/SSI applications and early retirement above either alone. Adding back pain to the illness definition eliminates the gender differential (men report more incident back pain), explaining some of the apparent sex differences in DI trends.
Depression × widowhood (indirect): Widowhood alone has a small impact on DI/SSI applications. Combined with baseline depression, the application probability rises approximately 4× over widowhood alone (Strategy 1). For men, this interaction is the largest single driver of early retirement — consistent with the widower mortality effect (Christakis and Iwashyna 2003) having a labor-market analog. The mechanism is bereavement as vulnerability amplifier, not a standalone trigger.
Measurement note: Conti et al. use Center for Epidemiologic Studies Depression Scale (CES-D) ≥4/8, which captures ~79% of Composite International Diagnostic Interview (CIDI) Diagnostic and Statistical Manual, Third Edition (DSM-III) major depression cases but overidentifies transient episodes — so the direct effect estimates are downward-biased. The true depression → DI/SSI application effect is likely larger than the 16% point estimate.
Reconciliation with cyclical DI literature: Depression explains secular DI level trends, not cyclical variation. Depression prevalence does not spike in recessions, so the Conti et al. finding is compatible with Cutler et al. (2012/2015) and Carey et al. (2025) rejecting health shocks as the mechanism for recession-induced DI surges. Both sets of findings can be correct simultaneously: mental health conditions enlarge the conditional applicant stock through a secular pathway while the entry-cost channel governs its recession-era activation.
Staubli (2011) provides international corroboration of the conditional applicant typology using Austria's 1996 DI reform. Austria's two-tier eligibility standard (any occupation below 55 / similar occupation above 55) creates an enrollment cliff at age 55 that concentrates marginal entrants — workers with moderate impairments and some employment alternatives — at exactly that threshold. The 1996 reform raised that cliff to 57 for men only, allowing a Difference-in-Differences (DiD) design (treated: men 55–56; comparison: men 49–54) with clean identification.
The heterogeneity results directly confirm the conditional applicant mechanism:
| Group | Pre-reform DI base | DI effect | Employment effect |
|---|---|---|---|
| Blue-collar | 33.3% | −8.9 to −11.2 pp | +2.7 to +6.0 pp |
| White-collar | 7.5% | −1.3 to −1.6 pp | null |
| Unhealthy | 43.5% | −9.7 to −12.4 pp | +3.3 to +7.1 pp |
| Healthy | 12.0% | −3.3 to −3.9 pp | null |
| Earnings Q1 | 43.4% | −12.0 pp | — |
| Earnings Q5 | 2.1% | −0.6 pp | — |
The pattern is exactly what the conditional applicant model predicts: workers at the eligibility cliff (high pre-reform DI base, blue-collar, low earnings) are disproportionately those with moderate impairments and employment alternatives. White-collar workers, who already faced a stricter own-occupation standard below 55, are not at the cliff in the same way — their marginal entrant population is smaller and less responsive.
The earnings gradient is explained by Austria's progressive benefit formula (lower earners have higher replacement rates → higher incentive to enroll near the eligibility threshold → larger response to a threshold change). This mirrors the DI Replacement Rate mechanism identified in the US context by Autor and Duggan (2003).
Comparison with Sweden (Karlström et al. 2008) and the US (Chen and van der Klaauw 2008): Austria's employment response (+1.6 to +3.4 pp) substantially exceeds both Sweden (no employment effect) and the US (moderate). Staubli attributes this to Austria's reform targeting a younger group (55–56 vs. 60+ in Sweden), with more residual employability — consistent with the conditional applicant model's prediction that responses are larger when the marginal enrollee has more viable employment alternatives.
Benítez-Silva, Buchinsky, Chan, Rust, and Sheidvasser (1999), using HRS waves 1–3 (n = 13,142), provide the earliest systematic evidence that a single self-reported indicator captures the dominant share of variation in DI application, appeal, and award decisions. Their key variable is HLIMPW: a binary indicator equal to 1 if the respondent reports "a health limitation that prevents them from working altogether."
HLIMPW has the largest coefficient and marginal effect at every decision node:
After conditioning on HLIMPW, most socio-economic predictors (race, marital status) lose significance. The authors interpret HLIMPW as a proxy for the individual's private information about true disability status — the same information that both drives application behavior and shapes government award decisions. The convergence of this single signal across all three decision nodes is consistent with a model where the individual's self-assessment is the closest available signal to the latent disability construct the DI program aims to measure.
The 30% incentive compatibility problem: Of DI applicants in the B-S sample, 30% report HLIMPW=0 — they do not consider themselves disabled. This group is concentrated in the lowest income decile, where DI's progressive benefit formula generates after-tax replacement rates exceeding 100%. This is the earliest direct HRS-based quantification of what Autor and Duggan (2003) later theorized as the incentive compatibility failure: at high enough replacement rates, application is rational even for individuals who do not meet the disability standard in any behavioral sense. See DI Replacement Rate.
Age-62 opportunistic application cliff: Application probability drops sharply for individuals aged 62+ (marginal effect −4.1 pp), despite DI benefits being approximately 20% higher than early retirement benefits and conferring Medicare eligibility up to one year earlier. The B-S interpretation: the hassle cost of the DI application process exceeds the expected utility of the extra 20% benefit margin for virtually all 62+ individuals. The option value of applying for DI collapses as soon as the early retirement alternative becomes available — applicants are rational in their comparison of DI versus Old Age and Survivors Insurance (OASI) early retirement, and the application cost (delay, documentation burden, uncertainty) tips the balance toward early retirement. This is the clearest pre-Autor-Duggan evidence of age-conditioned opportunistic behavior in DI application.
The conditional applicant model posits a stock of health-impaired workers who remain employed but whose work capacity is eroding — they are at risk of transitioning to DI application when job loss occurs. Rutledge, Zulkarnain, and King (2019) construct the closest observable measure of this latent stock at the occupation level: the Health Mismatch Index (HMI), which captures the share of workers who have a health limitation in an ability that their occupation requires.
At the cross-sectional level, HMI directly maps to Steps 4–5 of the SSA Sequential Determination Process: workers with high HMI are employed people whose health no longer fully satisfies their occupational requirements — exactly the conditional applicant type. Key findings for understanding the supply-side of DI applications:
The HMI's temporal trajectory during the DI growth period is particularly informative: it declined from 7.4% to 6.1% (1997–2010) — the same period in which DI rolls were growing substantially. This means the supply of health-mismatched workers (the latent conditional applicant pool) was shrinking, not growing, during the DI growth era. DI growth was driven by economic and demographic factors activating this pool faster than the pool itself was growing. The counterfactual: enrollment would have grown even more if health-job fit had not improved.
State-level analysis of vocational awards provides an independent fingerprint for the conditional applicant mechanism. The demographic group with the strongest relationship to vocational awards is workers aged 55–59 without a high school degree in manufacturing/production occupations — precisely the group most exposed to trade-related job displacement and long-term industrial decline. These workers drive both application rates and vocational acceptances but not medical-listing allowances. Michaud et al. also find that states with higher total application rates have higher vocational stage acceptance rates (not lower), suggesting that high-application environments pass a more marginal — not a looser-screened — pool to the vocational stage. Both patterns are consistent with conditional applicants: individuals with legitimate impairments who apply when labor market conditions make work untenable.
Coe and Rutledge (2013) provide a direct empirical portrait of how applicant characteristics shift across recession and expansion periods using two independent datasets: the Health and Retirement Study (HRS, waves 2000–2010, ages 50–Full Retirement Age (FRA)) and the SIPP Gold Standard File (2001 and 2004 panels, ages 25–61), both linked to SSA administrative records. Three comparison windows are used: the 2001–03 recession, the 2004–06 expansion, and the 2008–10 Great Recession.
Composition shift: Great Recession applicants are younger, better educated, higher income, and more likely to have recently held full-time jobs than expansion-period applicants. This is the demographic signature of conditional applicants — workers with qualifying impairments who were previously employed and apply only when job loss removes the employment alternative.
Mixed health evidence: SIPP data show GR applicants were 5.1 pp less likely to have had pre-recession work limitations (p<0.001), suggesting a healthier marginal applicant pool. HRS measures (ADLs, IADLs, CES-D depression score, mobility limitations) show no statistically significant differences — a discrepancy the authors attribute partly to HRS's older age range and smaller applicant sample.
Blinder-Oaxaca decomposition (Fairlie 2005 nonlinear extension): Observable characteristics explain fewer than 40% of the 0.7 pp rise in application rates (2.9%→3.6%) and fewer than 25% of the 9.1 pp rise in allowance rates (40.5%→49.6%). Over 60% of the application rate increase and 75% of the allowance increase are unexplained by observables — a large structural residual.
The award rate paradox: Given the composition shift toward healthier, more recently employed applicants, and the absence of programmatic changes during the 2000s, the DI award rate should have fallen. It rose from 1.2% to 1.8%. This implies a structural shift in SSA evaluation behavior — not just in the applicant pool — that the compositional story alone cannot explain.