A family of analytical frameworks for attributing changes in Social Security Disability Insurance (DI) enrollment to its component drivers — demographics, insured rates, incidence, mortality, and recovery. Four complementary frameworks exist, targeting different outcome variables and historical episodes:
The four frameworks answer different questions and produce different numerical results; none subsumes the others. Studies that analyze the beneficiary-to-population ratio (Autor and Duggan 2006; Burkhauser and Daly 2012) exclude population growth from the denominator by construction, making the incidence residual appear larger than the above frameworks find.
| Factor | Men | Women | Total |
|---|---|---|---|
| Population aging | 28% | 15% | 20% |
| Rising insured rate | 3% | 18% | 12% |
| Interaction (aging × incidence) | 4% | 19% | 13% |
| Rising incidence | 59% | 45% | 51% |
| Falling mortality | 8% | 3% | 5% |
| Recovery rates | −3% | 0% | −1% |
Caveat: The incidence figures use 1985 as a base. With a 1977 base (full 1977–2007 period): population aging 39%, insured rates 52%, incidence −10% for total — a near-complete reversal.
| Subperiod | Direction | Dominant Factors |
|---|---|---|
| 1977–1985 | Rolls fell | Falling incidence (64%), CDR-driven recovery (31%) |
| 1985–1993 | Rolls rose | Rising incidence (95% total); post-1984-reform bounce-back |
| 1993–2007 | Rolls rose (men) | Population aging (94%), falling mortality (36%); incidence below base (−23%) |
| 1993–2007 | Rolls rose (women) | Population aging (29%), rising insured rates (23%), rising incidence (38%) |
The 1985–1993 result for men is the most striking: rising incidence explains 125% of growth in the male beneficiary ratio (meaning all other factors, net, reduced enrollment). This is the policy-driven bounce-back from the 1984 reform — not a structural demographic shift.
Stable aggregate incidence post-1990 (after age and business-cycle adjustment) masks large offsetting trends by condition type:
Counterfactual: if musculoskeletal and mental incidence had stayed at 1985 levels (all other conditions following actual paths), the 2007 DI beneficiary ratio would be approximately 21% lower.
This compositional shift has two implications: (1) it mechanically reduces measured beneficiary mortality, independent of any genuine health improvement among beneficiaries; (2) it is consistent with the 1984 reforms making these categories easier to claim, amplified by declining relative wages for low-skill workers increasing application incentives.
Incidence measures the flow of new disabled-worker entitlements per year. Prevalence measures the stock on the rolls, which reflects incidence plus exit dynamics (mortality, recovery, OASI conversion at retirement age). Policy most directly controls entry rates; stock accumulation involves compounding effects over years or decades of changing composition. Neither measure is simply better — they answer different questions.
A related methodological trap: studies that study the ratio of beneficiaries to population (Autor and Duggan; Burkhauser and Daly) exclude population growth from the denominator. Because the working-age population grew over 1972–2008, its omission from the denominator inflates the apparent explanatory role of incidence. Pattison and Waldron decompose absolute entitlement growth and assign population growth its proper share.
Over 1972–2008, three demographic factors together explain 90% of the growth in new disabled-worker entitlements (from 254,200 to 897,000):
The adjusted disability incidence rate — the residual after controlling for all of the above — explains only 10%.
| Period | Demographic factors | Adjusted incidence rate |
|---|---|---|
| 1972–1990 | 83% | 17% |
| 1990–2008 | 94% | 6% |
| 1972–2008 | 90% | 10% |
In the 1990–2008 period — when unadjusted data suggest the steepest roll growth — the demographic explanation is stronger, not weaker.
The most important result is methodological. Unadjusted incidence rates appear to show a dramatic post-1990 acceleration:
| Period | Unadjusted incidence (avg annual) | Age/sex-adjusted incidence |
|---|---|---|
| 1972–1990 | −0.63%/yr | +0.27%/yr |
| 1990–2008 | +1.41%/yr | +0.15%/yr |
The unadjusted picture — a decline then a surge — is entirely an artifact of the baby boom. In 1972–1990, the baby boom was young (ages 8–44), contributing its population growth at low-incidence ages; this downweights incidence unadjusted. In 1990–2008, the baby boom moved into ages 44–62 — the peak disability-prone range — massively upweighting the unadjusted incidence rate. Once age/sex composition is held fixed, incidence grew more in the early period and slowed post-1990. The apparent post-1990 crisis in incidence is a statistical illusion.
The adjusted incidence rate is deliberately left as an unexplained residual. It absorbs: genuine health trends, economic conditions, changes in application behavior, perceived award probability, policy changes, legislative amendments, and judicial rulings. The large within-period fluctuations (peaks ≈1975, 1991, 2002; troughs ≈1982, 1997, 2006) are real policy- and cycle-driven effects. But at 10% of 36-year total growth, even if the residual were entirely attributable to program loosening, its quantitative significance for long-run program cost is modest relative to the demographic tide.
The composition shift from circulatory/cancer toward musculoskeletal/mental conditions documented by Liebman (2015) partly reflects a structural demand-side health trend: the doubling of U.S. adult obesity rates between the early 1980s and 2002. Obesity causally generates musculoskeletal conditions (low back pain, osteoarthritis), metabolic impairments (diabetes, cardiovascular precursors), and functional limitations — exactly the diagnostic categories driving the composition shift. Burkhauser and Cawley (2004) provide IV evidence that obesity raises disability income receipt by 5–9 percentage points (pp) (Two-Stage Least Squares [2SLS] using biological relatives' weight as instrumental variable [IV]), with Ordinary Least Squares (OLS) underestimating this effect by – due to measurement error and reverse causation. The parallel doubling of obesity rates and DI rolls since the early 1980s is consistent with obesity expanding the pool of health-impaired workers who are potential DI applicants. This is a health-demand contribution to the adjusted incidence residual that operates independently of the economic and policy incentive channels identified by Autor and Duggan. See Morbidity-Mortality Distinction and Conditional DI Applicants.
The Liebman (2015) decomposition isolates "falling mortality" as a distinct factor explaining 5–8% of DI growth (Table 1 above). Raut (2017) provides the microeconomic mechanism behind this channel and extends it to the recovery dimension. Using competing-risks analysis on 1981–2000 Continuous Work History Sample (CWHS) entrants, Raut shows:
The death-exit decline: Death exit rates declined for all subgroups between the 1981–1985 and 1991–1995 entrant cohorts — across all age groups, both sexes, and all impairment types. Beneficiaries are surviving longer on the rolls.
The recovery stagnation: Recovery exit rates were largely unchanged between the same cohorts, with modest improvement only for ages 41–50. Recovery — at approximately 8% over 30 years for the full population — was already rare; it did not improve as medicine extended survival.
The interaction of these two trends is the duration channel: when death exit rates fall but recovery exit rates do not rise, average spell lengths increase. A fixed annual incidence rate therefore produces a growing stock of beneficiaries. This mechanism is distinct from the incidence channel (new entrants per year) and can produce roll growth even when annual application rates are stable.
The female duration effect: Women entering the DI rolls have lower death exit rates AND lower recovery exit rates than men with comparable impairments and ages at entitlement. At ages 20–30 with musculoskeletal disorders, men's 9-year recovery probability is 38.5% vs. women's 20.3%. This double-low structure generates longer female average spells, amplifying the impact of the rise in female DI insured rates documented by Pattison and Waldron (2013): not only did more women enter the insured pool, but each female entrant stayed longer on average.
The composition interaction: The same compositional shift from high-mortality conditions (circulatory, neoplasms) to low-mortality conditions (musculoskeletal, mental disorders) that reduced measured beneficiary mortality also extended average spell durations — because musculoskeletal/mental entrants die more slowly, contributing to roll accumulation even absent any change in incidence or recovery rates. Liebman's 21% counterfactual captures the incidence side of this shift; the Raut findings capture the survival/duration side.
A complementary decomposition focuses specifically on the vocational stage, adding education and occupation to the age-only frameworks above. Using state-level data (2001–2015), Michaud, Nelson, and Wiczer find that demographic composition explains of the rise in vocational awards and of total determinations. Population aging increases awards; rising educational attainment more than cancels it; occupational shifts (manufacturing decline) also work against award growth. Taken together, observable demographic change predicts that vocational awards should have fallen — yet they rose threefold since 1985. The majority of the trend must reflect behavioral change (application propensity within demographic cells) or implementation drift in how SSA applies the vocational grid rules. This is consistent with the Liebman (2015) finding that adjusted incidence — not demographics — drove the 1985–1993 surge, but it extends that puzzle specifically to the vocational stage.
The decomposition frameworks above treat the application rate as the key behavioral margin — but a competing hypothesis is that working-age Americans became genuinely sicker, making DI growth a medically appropriate response rather than an economic one. Rutledge, Wu, Guan, and Trenkamp (2014) provide the most direct test using Survey of Income and Program Participation (SIPP) panels (1990–2008) linked to SSA 831 File administrative data, measuring the pre-application health of approximately 3,250 Social Security Disability Insurance (SSDI)/Supplemental Security Income (SSI) applicants 1–3 years before filing (to avoid strategic reporting bias).
Key findings:
The allowed/denied comparison is revealing: awardees' ADL/IADL count worsened over time while denied applicants' health improved relative to awardees. The eligibility process was selecting a progressively sicker awardee subset from a stable (or improving) overall applicant health distribution — consistent with a marginal applicant pool expanding from the economic side without a parallel health decline.
This evidence rules out health deterioration as a primary driver of DI roll growth and corroborates Pattison and Waldron's finding that the adjusted incidence residual is small. The growth in applications came from economic and demographic push, not from a wave of sicker workers. See Conditional DI Applicants.
Liebman's decomposition captures structural growth — long-run changes in demographics, insured rates, and incidence trends. It does not capture the cyclical dimension: at the business-cycle frequency, recessions activate the existing stock of conditional applicants (people with qualifying impairments who prefer work when employed but apply after job loss), producing surge-and-retreat patterns that are layered on top of the structural trend. Lindner, Burdick, and Meseguer (2017) document this cyclical mechanism, finding that a 1 pp rise in unemployment increases applications ~3.1% — concentrated at Steps 2 and 4 of the determination process, not at Step 3 (Listing-eligible; severe, inframarginal cases). See Conditional DI Applicants.
A prominent hypothesis for SSDI's countercyclical pattern is that displaced workers exhaust Unemployment Insurance (UI) benefits and then apply for SSDI — the "UI-before-SSDI" mechanism. Mueller et al. (2016) directly test this using the Great Recession's dramatic variation in UI extension duration (26 to 99 weeks across states and time), a powerful natural experiment. Their findings:
The null result is consistent with the Michaud-Wiczer (2018) structural finding that business cycles have near-zero impact on SSDI awards (despite large application effects): the cycling-in of conditional applicants changes composition, not total awards, and is not driven by the UI-exhaustion channel.
Autor and Duggan (2006) synthesize the post-1984 expansion into an explicit three-cause framework, adding diagnostic composition data and fiscal projections to their 2003 analysis.
The three causes:
What the 2006 paper rules out: Population aging explains at most 6% of growth (not 40–90% as later demographic frameworks find for different periods). Population health improved over the same period — mortality among 50–64-year-olds fell 17–29% (1981–2001), ruling out health deterioration as a driver.
Fiscal projections (2006 paper): DI spending as share of total SS outlays: 10% (1985) → 17% (2005). DI payroll tax: 1.0% → 1.8% of covered wages. Average new award present value (cash + Medicare): $245,000. Projected steady-state: 7% of nonelderly adults from 4.1% in 2005. This projection did not materialize — enrollment peaked ~2014–15 and declined — because the framework understated the baby boom's natural demographic ceiling and the labor market's recovery capacity. See Conditional DI Applicants for the Deshpande et al. (2025) post-2010 decline analysis.
Autor and Duggan (2003) offer the most influential alternative explanation for DI growth: a rising earnings replacement rate driven by the progressive benefit formula interacting with declining real wages of low-skill workers. As wages at the bottom fell while the SSA national wage index (used to compute benefits) rose, the DI-to-earnings ratio mechanically climbed for low-wage workers. The demand-shock IV evidence (Bartik industry-mix instrument) shows post-1984 application rates responding – more strongly to labor demand shocks, attributed to this replacement rate effect. See DI Replacement Rate.
This framework has three important limitations relative to Pattison-Waldron and Liebman:
What survives: the replacement rate is a real explanation for who expanded (low-education, low-wage workers) and why the diagnostic mix shifted toward subjective conditions. It is less defensible as an explanation for the aggregate magnitude of program growth.
The permanent-inequality premise confirmed (DeBacker et al. 2013): A necessary condition for the Autor-Duggan replacement rate mechanism to have lasting effects on application incentives is that the rise in earnings inequality reflects a durable structural shift rather than temporary volatility. DeBacker, Heim, Panousi, Ramnath, and Vidangos (2013) directly test this using a confidential IRS panel (1987–2009): they find that the entire rise in male earnings inequality is attributable to the permanent variance component, with zero increase in transitory variance. For household income, approximately 75% of the inequality increase is permanent. This confirms that the low-skill wage stagnation underlying the replacement rate channel is a structural change that would persistently raise DI incentives for the marginal worker — not a cyclical fluctuation that would reverse. See Income Dynamics.
Duggan and Imberman (2009) construct the first explicit percentage decomposition of DI recipiency growth from 1984 to 2003, using sequential counterfactual simulations, cohort-tracking, and time-series regressions. Each factor is estimated separately under an orthogonality assumption; interaction effects are not modeled. Results are reported by sex because the dominant factors differ across groups.
| Factor | Women | Men |
|---|---|---|
| Age structure (baby boom aging) | 4% | 15% |
| DI-insured status (rising female LFP) | 24% | 3% |
| Economic conditions (1991 + 2001 recessions) | 12% | 24% |
| Replacement rates (inequality × progressive formula) | 28% | 24% |
| Medical eligibility criteria (1984 DBRA) | 38% | 53% |
| Total explained | 106% | 119% |
Over-explanation (>100%) is informative: the totals exceed 100% because health improvements among the near-elderly (National Health Interview Survey [NHIS] 1984–2002: lower self-reported work limitations and inability to work) actually reduced DI growth below what it otherwise would have been. Absent these health improvements, rolls would have grown by an additional ~6–19% — health is effectively a negative factor not shown in the table.
Medical eligibility criteria as the dominant driver: The DBRA 1984 tripled mental disorder and musculoskeletal award rates per 1,000 insured persons (from 0.88 to 2.67 over 1983–2003) while cancer/circulatory awards were essentially flat (1.15 → 1.12). The Duggan-Imberman identification assumption is that the "other conditions" award rate trajectory is a valid counterfactual for what mental/musculoskeletal (MSK) would have been without the eligibility change. This contrasts with Ruffing's (2014) SSA actuarial estimate that DBRA's long-run cost impact was negligible — the discrepancy arises because Ruffing's estimate reflects the legislation's narrowly defined provisions, while Duggan-Imberman capture all award growth correlated with the post-DBRA mental/MSK surge.
Declining beneficiary exit rates as a structural amplifier: Annual exit rate fell from 14.4% (1984) to 7.9% (2003), driven by the composition shift toward longer-duration conditions. This implies the program was far below its equilibrium stock in 2003: at 2003 award and exit rates, equilibrium ≈ 9.8M recipients (vs. 6.2M then), predicting continued roll growth.
Comparison with later frameworks: Duggan-Imberman is best read as a complement to Pattison-Waldron (2013) and Liebman (2015). Where those frameworks emphasize demographic adjustment and sequential counterfactuals on long time series, Duggan-Imberman gives a gross attribution across all five drivers simultaneously, including economic conditions explicitly. The frameworks broadly agree on the non-demographic drivers' importance but differ in magnitudes because of different study periods, outcome variables, and attribution methods (see the "Cross-Study Reconciliation" subsection below).
Duggan and Imberman's attribution of ~50% of DI roll growth to the 1984 DBRA is a quasi-experimental decomposition: other-condition award rates serve as the counterfactual for what mental/MSK awards would have been without the eligibility change. Duggan, Rosenheck, and Singleton (2009) provide a sharper causal test of the same medical eligibility mechanism using the Veterans Affairs (VA) Disability Compensation (DC) program.
The July 2001 Agent Orange decision — deeming type II diabetes service-connected for Vietnam theater veterans but not for other veterans — is an unusually clean natural experiment: a specific date, a specific condition, and an identifiable treatment group with clear pre-policy parallel trends. The difference-in-differences estimate shows the policy increased DC enrollment by 6 percentage points among Vietnam theater veterans (200–238K new beneficiaries) and DC expenditures by 50B present value). Roughly 8% of potentially affected veterans responded to the eligibility expansion — a magnitude consistent with what Duggan-Imberman's 53%/38% figures imply about the aggregate SSDI response to the broader 1984 DBRA changes. The Agent Orange result directly validates the identifying assumption in the Duggan-Imberman decomposition: when medical eligibility loosens for a defined group, that group's enrollment accelerates, not some confounding trend.
A secondary finding — no labor supply effect for the veterans themselves, but significant reduction in wives' labor supply — suggests the behavioral response to disability income works largely through household wealth effects rather than individual work disincentives in settings (like the VA DC program) that allow beneficiaries to work. This contrasts with the stronger labor supply effects found for SSDI/SSI, which create implicit taxes on earnings.
Ruffing (2014) performs a count-based accounting of DI growth that directly assigns beneficiary totals to five demographic and policy factors and uses the comparison to reconcile the apparently contradictory results across the literature. Unlike the rate-based frameworks above, it counts absolute beneficiaries — so overall population growth enters as a positive demographic contributor rather than being absorbed into the denominator.
The Five Factors and Their Magnitudes
Between 1980 and 2013, DI disabled-worker rolls grew from 2.9 million to 8.9 million (a 6.0 million increase). Five identifiable factors account for million of that growth:
| Factor | Additional beneficiaries | Mechanism |
|---|---|---|
| Overall population growth | +1.25 million | Working-age population grew |
| Population aging | +900,000 | Baby boomers entering peak disability-prone years (55–64) |
| Women's LFP rise | +900,000 | Insured women rose from to of workforce |
| FRA rise () | +450,000 | Workers who would have converted to OASI at 65 remain on DI rolls an extra year |
| Women's catch-up | +650,000 | Insured women went from as likely as insured men to receive DI (1980) to virtual parity (2013) |
Total: million → 78% of the 2013 enrolled total (8.9M actual vs. M from these five factors) and 68% of growth since 1980.
The FRA Effect
The FRA rise from 65 to 66 — phased in for 1938–1942 birth cohorts, fully effective for those born 1943–1954 — is a mechanically pure policy factor: workers who previously converted to OASI at 65 remain on DI rolls for an extra year. Ruffing estimates this accounts for more than 5% of recent DI enrollment. The subsequent FRA rise from 66 to 67 (phased in for 1955–1960 birth cohorts, 2017–2022) will add further cost pressure. Neither Pattison-Waldron (through 2008) nor Liebman (through 2007) fully capture this effect.
Ruffing's mechanical FRA effect (+450,000 beneficiaries who remain on DI rolls rather than converting to Old-Age (OA) benefits at 65) is a duration channel: the FRA rise delays the exit of existing beneficiaries. Duggan, Singleton, and Song (2007) identify a distinct entry channel operating at the application margin: the same FRA rise that delays OA conversion also reduces the generosity of early retirement benefits at age 62, raising the financial incentive to apply for DI rather than claim early OA.
The mechanism: Early OA benefit at 62 equals 80% of Primary Insurance Amount (PIA) for workers born ≤1937. The 1983 legislation phases this to 75% (born 1943–1945) and ultimately 70% (born 1960+). DI benefits are unchanged. The expected present-value advantage of DI over early OA therefore rises from ~45,991 (born 1943) to ~$55,190 (born 1960). Using SSA Continuous Work History Sample data for men born 1935–1945 and year-of-birth variation in the OA/DI benefit ratio as a natural experiment:
The two FRA channels are additive and distinct: Ruffing's mechanical effect operates through the exit (duration) margin; Duggan-Singleton-Song's behavioral effect operates through the entry margin. Together they quantify the full DI cost of the 1983 FRA reform. The small magnitude of the behavioral channel (~42K, or 1.3% of male rolls) is itself a key finding: the substitution incentive is real but modest, and does not undermine the fiscal case for raising the FRA.
Women's Catch-Up: An Under-Studied Factor
The convergence of female to male DI receipt rates within the insured population — from as likely to virtually equal — added 650,000 beneficiaries and is absent from most prior analyses. It is distinct from the rise in insured rates (more women qualifying for coverage, captured by Pattison-Waldron and Liebman). The mechanism remains under-studied: candidates include occupational convergence into physically demanding or high-stress jobs, reduced option value of informal caregiving, or genuine health convergence as women accumulated more decades of full-time employment. Ruffing does not identify a causal explanation.
The DBRA Counter-Evidence
A common narrative attributes the DI boom partly to DBRA 1984 (which relaxed the medical improvement standard for CDR terminations and gave greater weight to pain). Ruffing cites SSA's own actuarial estimate that DBRA's long-run financial impact was approximately of taxable payroll — effectively negligible. This supports the Liebman (2015) interpretation that the 1985–1993 incidence surge reflects a return from the artificially compressed CDR-era trough, not a policy-induced structural expansion.
Cross-Study Reconciliation
Literature-wide demographic attribution estimates range from 40% (Autor-Duggan) to 90% (Pattison-Waldron). Ruffing's Table 1 shows the discrepancy dissolves once three measurement choices are controlled:
| Study | Outcome | Period | Women's catch-up | Demographic share |
|---|---|---|---|---|
| Ruffing (2014) | Number | 1980–2013 | Yes | 78% (68% of growth) |
| FRBSF | Rate | 1980–2011 | No | 56% |
| CBPP prior | Rate | 1989–2011 | No | 48% |
| Autor-Duggan | Rate | 1989–2011 | No | 40% |
| Pattison-Waldron | Incidence | 1972–2008 | Partial | 90% |
These are consistent answers to different questions, not contradictory findings.
Non-Demographic Remainder
The remaining 22% of the 2013 total reflects non-demographic forces: workplace changes (globalization, technology, declining accommodation of impaired workers), the Medicare 2-year wait creating an enrollment retention motive, falling beneficiary death rates (/yr) extending average spell length, underfunded CDRs, and the Great Recession's cyclical boost to applications. DI costs peaked at of GDP (2010–2013) and are projected to stabilize near as baby boomers age off DI rolls onto retirement rolls.
Michaud and Wiczer (2018) complement the reduced-form frameworks above with a quantitative structural model in which SSDI application is an "option" exercised by long-term non-employed workers. The model features 16 occupations, 5 age groups (30–64), and 3 health states, calibrated to O*NET, Panel Study of Income Dynamics (PSID), and Current Population Survey (CPS) data. Unlike the accounting frameworks, it can assign causal weights to simultaneously operating forces using a Shapley-Owen decomposition of simulated SSDI new awards.
Decomposition results:
| Driver | Pre-2000 contribution | Post-2000 contribution |
|---|---|---|
| Secular wage declines | 24% | smaller |
| Baby-boom aging | 13% (smaller) | 13% (dominant) |
| Business cycles |
The near-zero business cycle contribution to awards is the paper's sharpest finding: cycles affect applications substantially (job loss elasticity ≈ 0.07–0.17) but leave awards nearly unchanged, consistent with compositional shifts among applicants rather than screening standard changes. This separates cyclical application surges (documented in Conditional DI Applicants) from the structural drivers of long-run award growth.
Occupational bundling: The structural mechanism linking wage trends to SSDI is occupational. Occupations with a 5 pp higher work-limitation hazard by age 60 have on average a 1% lower secular wage trend. Physical occupations therefore face double exposure: accelerated health deterioration and wage stagnation simultaneously make SSDI more attractive. This explains why the application elasticity to secular wage decline is for health-impaired workers () vs. for healthy workers () — impaired workers in declining-wage occupations are the marginal group driving structural award growth.
The SSDI non-employment shadow: Workers in the application pipeline (5-month non-employment requirement before filing) constitute 1.6–2.5% of the working-age population at any time. Non-employed applicants grew from 25% of all non-employed workers (1985) to 66% (2010). Omitting applicants from employment counts understates SSDI's measured labor market footprint by 30–40%.
The model replicates approximately two-thirds of the 6.4 pp male non-employment increase attributable to SSDI over 1985–2010 (predicting 4.3 pp): 75% via current beneficiaries, 18% via applicants in the pipeline. See Conditional DI Applicants for the non-employment shadow implications.
Deshpande, Kellogg, Mogstad, and Tseng (2025) embed the enrollment accounting identity
where = enrollees, = entry rate, = eligible population in group , and = exit rate — in a partial-identification framework. The five decision margins (population composition, eligibility, application, award, exit) interact: a change in population size simultaneously affects the eligible pool, the applicant pool, and the award stage. Standard Oaxaca-Blinder sequential decomposition assigns interaction terms to whichever margin is evaluated first; the bounds approach instead reports intervals for each margin's contribution that are valid regardless of the interaction attribution rule.
The application margin is the dominant factor explaining enrollment growth over this period. Demographic factors — population growth and aging into peak disability-risk ages — played a secondary role. Award rates and exit rates contributed modestly in the expected direction but were not the primary drivers.
| Margin | Naive share | Composition-adjusted share |
|---|---|---|
| Application decline | ||
| Award rate decline | ||
| Exit rate change | small | small |
The naive decomposition overstates the award rate's role. As the labor market recovered post-2010, healthier conditional applicants returned to employment, leaving a compositionally more-severe applicant pool. This mechanically reduced award rates — not because eligibility standards tightened, but because the pool that did apply was more severe. The composition-adjusted figures correct for this, attributing ~71% of the decline to the application margin.
Labor demand (dominant) Using a Bartik shift-share IV (Autor-Duggan design) exploiting variation in local industry employment mix:
ALJ reform (2017–2019) Award rates at the Administrative Law Judge (ALJ) stage fell sharply under policy changes during 2017–2019. If applicants rationally anticipate their probability of ultimate award, this should reduce marginal applications. The paper finds applications did not respond to the ALJ award rate decline — the elasticity of applications to ALJ award rates is approximately zero. ALJ reform accounts for at most 0–3% of the total enrollment decline. This falsifies the rational-application hypothesis for the ALJ stage; potential applicants either do not observe ALJ award rates, discount them heavily relative to initial-stage expectations, or have already filtered on the initial-stage decision.
Field office closures SSA field office closures over 2010–2019 account for approximately 3–6% of the enrollment decline — consistent with the Deshpande-Li (2019) finding that closures reduce applications, but small relative to the labor demand channel.
Population health Disability rates in the general working-age population did not improve sufficiently between 2010 and 2019 to explain the observed enrollment decline. Population health changes cannot account for the timing or magnitude of the fall.
The decline is concentrated among workers aged 50–64, workers with lower education, and applicants with musculoskeletal disorders — precisely the demographic profile of conditional applicants who prefer work when employment is available. This heterogeneity pattern is consistent with labor demand recovery selectively drawing the most work-capable health-impaired individuals back into employment, leaving the most severely disabled as the residual applicant pool.
Because ~71% of the decline reflects the application margin, and ~73% of that is driven by labor demand recovery, most declined enrollees are people who chose to work rather than being screened out by tighter eligibility. From a social insurance standpoint, this is different from a policy-induced decline: the welfare cost of deterring eligible individuals is absent when those individuals voluntarily prefer employment. However, the decline raises two concerns: (1) the remaining applicant pool is more severely disabled and may have unmet needs; (2) the decline is mechanically reversible — a future recession would likely activate the conditional applicant stock again, producing a surge comparable in structure to the 2008–2010 wave.
The decomposition resolves the Autor-Duggan / Reno debate by showing both are right about different periods. The "fiscal crisis" narrative correctly identified the 1980s policy bounce-back; the "demographic forces" narrative correctly characterizes the post-1992 period. Conflating the two — which is easy given the monotonic aggregate enrollment trend — produces misdiagnosed reform prescriptions.
If post-1992 growth is primarily demographic (baby boomers aging into peak disability years, women entering the insured pool), then CDR crackdowns and eligibility tightening are poorly targeted remedies. Liebman (2015) argues that "all three factors have now arguably run their course" as of around 2007 — the aging wave, the female insured-rate convergence, and the post-1984 incidence rebound have all peaked.
If DI enrollment growth was driven partly by worsening health-job fit — workers increasingly unable to meet the physical and cognitive demands of their occupations — then a measure of that mismatch should have risen alongside the rolls. Rutledge, Zulkarnain, and King (2019) test this directly using the Health Mismatch Index (HMI): the share of workers in each occupation reporting a health limitation in an Occupational Requirements Survey (ORS)-required ability, constructed from SIPP panels (1997–2010).
The key finding is the opposite of what a health-deterioration story would predict: the HMI fell from 7.4% (1997) to 6.1% (2010) — a 1.3 pp decline — despite workforce aging and growing DI rolls. The decline is concentrated in the 1-mismatch group; the 2+ and 3+ shares were roughly stable. This means:
The HMI is also predictive of SSDI entry: pp HMI pp of an occupation's workforce entering SSDI in the next 16 months (nearly doubling the mean). The effect is concentrated in workers 50–64, consistent with the Vocational Grid's age thresholds and the conditional applicant framework. This establishes that health-job mismatch is a channel through which occupational structure feeds the DI pipeline — but one whose contribution was declining in the period of maximum DI growth.