Vocational Grid

Definition

The Social Security Administration (SSA)'s medical-vocational guidelines (informally, "the vocational grid"), formally introduced in 1978, provide a decisional framework for disability insurance (DI) eligibility when a claimant's impairment is severe but does not meet SSA's medical Listing of Impairments. The grid combines residual functional capacity (RFC), age, education, past work experience, and transferable skills to determine whether a claimant can perform any work in the national economy. Key age thresholds — particularly at 50 and 55 — relax eligibility requirements and produce discontinuous jumps in allowance rates.

Key Ideas

Age is treated as a vocational factor because older workers are assumed to have greater difficulty adapting to new jobs and learning new skills — not because they have fewer remaining years of life.
The grid creates economic incentives at threshold ages: allowance rates jump sharply at 50 and 55.
Despite this, claiming behavior does not cluster around threshold ages — applicants do not appear to time applications strategically.
A "borderline age policy" permits disability examiners to use the next age category if the claimant is within 6 months of the threshold and that age would result in allowance.

How It Works — Mortality Signature

The vocational grid leaves a visible fingerprint in DI mortality curves. Because threshold ages allow entry to claimants who are medically less severely impaired (they qualify on vocational grounds rather than meeting a listing), beneficiaries who enter the rolls at exactly age 50 or 55 have lower mortality than those who entered at ages 49 or 54.

In 2011–2015, probability of death at first year of entitlement: age 49 $= 0.062$ ; age 50 $= 0.043$ (a $30\%$ drop).
The same male beneficiary faced a mortality rate at age 60 (first year) similar to age 49 — illustrating the overall flattening of the DI mortality curve at older ages.
These dips were imperceptible in the 1970s, moderate in the early 1990s, and pronounced by 2011–2015 — directly tracking the rising share of vocational-based awards ( $24.3\%$ in 1992 $\to$ $40.1\%$ in 2015 $\to$ $\approx 49\%$ in 2018).
The lower mortality among threshold entrants persists for multiple years of duration, not just year 0.

Effect on Life Expectancy

A man entering at age 50 in 2011–2015 had a remaining life expectancy of $19.48$ years — about $0.72$ years higher than a linear extrapolation from ages 49 and 51 would predict. For women, the corresponding gain was $\approx 0.77$ years. The vocational grid thus modestly but measurably inflates period life expectancy at those ages.

Empirical Performance of the Grid (Michaud et al. 2017)

The first systematic test of whether vocational grid rules operate as designed uses state-level data (2001–2015) to regress Social Security Disability Insurance (SSDI) vocational awards and denials on state demographic composition (age × education × occupation). Key results:

Education works as designed: high school graduates have lower vocational acceptance rates than non-graduates, consistent with the grid's intent that more education expands available job options.
The 55–59 age group, not 60–64, drives awards. Despite the grid being most generous at age 60+, workers aged 55–59 produce the largest total effect on both applications and vocational acceptances — because of their greater population size and much higher application propensity.
Demographic composition explains $\leq 16.5\%$ of the trend in vocational awards. Population aging increases awards; rising educational attainment more than offsets it. The majority of the $3$ – $4\times$ rise in vocational awards since 1985 cannot be attributed to the observable demographic characteristics the grid conditions on.
A puzzle: the conditional acceptance rate at the vocational stage roughly doubled from 1985 to 2000 despite rising education — the opposite of what the grid's design predicts. Whether this reflects implementation drift, unobserved health deterioration in the 55–59 manufacturing demographic, or selection is unresolved.
Cross-state variation: $61\%$ of state-level variation in vocational awards is driven by variation in total applications; $35\%$ by the conditional acceptance rate at the vocational stage. High-application states (South, Rust Belt) screen harder at earlier steps, passing a more marginal pool to vocational evaluation.

Causal Identification: Grid Cutoffs as an RD Instrument (Chen and van der Klaauw 2008)

The age thresholds at 45, 50, and 55 do more than create a mortality fingerprint — they provide a source of quasi-random variation in DI award probability that can identify the causal work-disincentive effect of program receipt.

Chen and van der Klaauw (2008) exploit this with a fuzzy regression-discontinuity (RD) design for Stage 5 (vocational) applicants. The design is "fuzzy" rather than "sharp" because the grid is not mechanically deterministic at the cutoffs: unobserved factors — residual functional capacity, illiteracy, and examiner discretion — also influence the award decision. As a result, award probability is discontinuous at ages 45, 50, and 55 but not a step function. Applicants just below a cutoff (who face a discontinuously lower award probability) serve as counterfactual for those just above (who are discontinuously more likely to be awarded).

Key results:

Work-disincentive effects for Stage 5 vocational applicants near a cutoff are smaller than the comparison-group upper bound for all applicants ( $\approx 11$ – $16$ percentage points (pp) vs. $\approx 20$ pp), consistent with the interpretation that vocational-stage applicants retain more residual work capacity
SSDI male applicants show somewhat larger responses than female or Supplemental Security Income (SSI) applicants within this subgroup
The design applies only to the $\approx 37$ – $40\%$ of applicants who reach Stage 5 — the subset for whom the grid is relevant

This RD approach is a direct methodological precursor to the examiner/judge instrumental variable (IV) designs of Maestas, Mullen, and Strand (MMS, 2013) and French and Song (2014), which generalize quasi-random assignment to all DI applicants rather than the Stage 5 vocational subset.

Extension to financial distress outcomes (Deshpande, Gross, and Su 2021): The same age cutoffs are used to identify the causal effect of DI allowance on bankruptcy, foreclosure, home sales, and home purchases — the first quasi-experimental evidence on the benefit side of DI. The paper augments the basic RD with an "office classification" strategy that exploits heterogeneity across $\approx 130$ Disability Determination Services (DDS) offices in applying the borderline age rule. "RD offices" ignore the rule entirely (producing a clean RD at the cutoff); "Spline offices" fully implement it (producing only trend breaks, no jump); "Hybrid offices" partially implement it. This three-way classification yields more precise IV estimates than the standard RD alone by recovering additional variation from the pre-cutoff trend breaks in Spline and Hybrid offices. All three strategies produce near-identical point estimates, confirming robustness. See Nonhealth Risk and DI Insurance Value for how these estimates connect to the welfare decomposition.

Age as Dominant Allowance Factor: Descriptive Confirmation (Contreary et al. 2017)

Contreary, Honeycutt, Stegman Bailey, and Mastrianni (2017) provide a descriptive corroboration of the vocational grid's age dominance using Survey of Income and Program Participation (SIPP)-SSA linked data for DI applicants ages 25–55. In a regression predicting allowance rates with demographic and employment history controls, age is by far the largest single predictor: applicants ages 51–55 have a $+30$ pp higher allowance rate than applicants ages 25–30. This dwarfs the effects of pre-application employment pattern (consistently employed vs. no employment: $\approx 10$ – $15$ pp difference) and most other covariates. The result confirms that the grid's age thresholds operate as designed — generating a steep gradient across the 25–55 age range well before the formal 55-threshold jump — and that early-intervention proposals focused on employment recency are working against the single largest determinant of DI awards.

Workers' compensation (WC) receipt at the 6-month pre-application window carries a $-13$ pp allowance penalty. This is plausible within the vocational grid framework: WC receipt signals that the applicant recently engaged in (and was injured during) substantial work activity, which SSA may read as evidence of residual work capacity relevant to Steps 4–5 of the determination process.

The Step 5 Denied Population (Strand and Trenkamp 2015)

Claimants who pass Step 4 but are denied at Step 5 — those with residual capacity for some work in the national economy — represent the own-occupation impairment group that DI explicitly excludes by design. Strand and Trenkamp (2015) study $37{,}110$ such claimants denied in 2005 who did not appeal or reapply. Their post-denial labor market outcomes quantify the "coverage gap":

Employment fell from $85\%$ to $63\%$ ( $-21.8$ pp)
Median conditional earnings fell to $77$ cents on the dollar
$78\%$ worked at some point over the subsequent six years — similar to vocational rehabilitation (VR) recipients, far above allowed DI beneficiaries

The diagnosis-level breakdown is directly interpretable through the vocational grid logic: mental disorders, where functional impairment is harder to observe and less tied to physical job requirements, produce the largest post-denial employment losses. Sensory and cardiovascular/neurological conditions — where the occupation-change constraint is more tractable — produce the smallest losses. This suggests the grid's implicit model that denied claimants can transition to other work is better supported for some diagnostic categories than others.

The Vocational Awards Puzzle (Michaud and Wiczer 2018)

Michaud and Wiczer (2018) provide a structural quantification of a puzzle implicit in Michaud et al. (2017): the vocational award share rose from $\approx 25\%$ (early 1990s) to $\approx 60\%$ (post-2010), a $35$ pp increase. Their overlapping generations (OLG) model — calibrated to the vocational grid's documented age, education, and occupation rules — holds those rules constant and simulates what vocational award shares should have been under stable implementation.

The finding: Constant rules predict only a $\approx 6$ pp rise in the vocational award share. The actual data show a $35$ pp rise. The gap of $\approx 29$ pp is not explained by the demographics and occupational composition the model captures.

Interpretation: The gap implies that either (a) there has been substantial de facto implementation drift — SSA examiners and administrative law judges (ALJs) applied the vocational grid more generously over time, even without formal rule changes — or (b) there has been a large unobserved health deterioration among the 50–64 low-education manufacturing demographic that the model's health-state transitions do not capture. The paper's own calibration evidence (drawn from Panel Study of Income Dynamics (PSID) health transitions) argues against the latter.

This structural evidence complements the empirical evidence for examiner/ALJ leniency variation documented in French and Song (2014) (ALJ lottery IV) and the award-rate decline post-2010 documented in Deshpande et al. (2025) — both consistent with discretionary implementation, not purely mechanical rule application.

Implication for policy: If the $35$ pp vocational award share rise reflects implementation drift rather than genuine rule changes or population health decline, it suggests that published DI growth decompositions that take award rates as rule-determined may be misattributing implementation-driven award rate increases to health or demographic factors. The puzzle is unresolved.

Geographic Expression of the Grid (Ruffing 2015)

The education–disability receipt correlation documented across U.S. states is the geographic expression of the vocational grid mechanism operating at scale. Ruffing (2015) shows that four factors explain approximately $85\%$ of cross-state variation in DI/SSI receipt rates (ordinary least squares (OLS), 2013 data): low educational attainment, older workforce, fewer immigrants, and blue-collar industry mix. Education is by far the strongest predictor.

The mechanism is direct: SSA's statutory framework (Social Security Act §223) requires applicants to show not just that they cannot do past work but that they cannot perform any work in the national economy — an adjustment that depends critically on education and skill transferability. A college-educated professional with a serious condition typically has multiple retraining options; a 60-year-old high-school dropout in a manual occupation with the same condition faces no realistic alternatives under the "any work" standard. The grid codifies this distinction through its age-bracket and education rules, and the geographic pattern is the aggregate result of those rules applied to heterogeneous state workforces.

High-receipt states (WV $12.5\%$ , KY $11.4\%$ , AL $11.0\%$ ) combine all four factors: older, less-educated, blue-collar-intensive workforces with few immigrants. Low-receipt states (UT $3.8\%$ , HI $4.1\%$ ) have younger, more-educated, service-sector workforces with higher immigrant shares. Unemployment rates explain almost none of the geographic variation — because although recessions activate applications at Steps 2 and 4, the grid's vocational standards do not loosen during downturns; marginal recession-era applicants are more likely to be denied. The geographic distribution is structural, not cyclical.

Within-state variation can rival between-state variation: Virginia simultaneously held five of the ten highest-rate U.S. counties and four of the ten lowest (the DC suburbs) in 2011, reflecting intrastate heterogeneity in education, age, and industry mix that maps directly onto the grid's eligibility criteria.

The Grid Reform Debate: 2005 NPRM, 2009 Withdrawal, and 2015 ANPRM

SSA first proposed revising the vocational guidelines in 2005, when it issued a Notice of Proposed Rulemaking (NPRM) proposing to add 2 years to all grid age thresholds — e.g., the criteria applying to workers aged 50–54 would shift to 52–56. SSA's stated rationale was that "advances in medical treatment and healthcare have provided longer life expectancies and more healthy years for millions of Americans." The rule was withdrawn in May 2009 after substantial negative public comments. SSA re-opened the question in September 2015 via an Advance Notice of Proposed Rulemaking (ANPRM) on vocational factors.

Van de Water (2015) offers the canonical rebuttal to the 2005/2015 rationale:

Averages are misleading when health varies by socioeconomic status (SES). Studies finding that Americans on average are healthier at a given age than before "are not relevant, however, to those individuals who are not healthy at these ages." Health improvements and life expectancy gains since the 1990s have been concentrated among high-earners: the gap in life expectancy at age 65 between the top and bottom halves of male earners widened by $5.4$ years between 1977 and 2006 (Waldron 2007). College degree holders delay onset of poor health by $\approx 20$ years relative to high school (HS) graduates, $\approx 30$ – $40$ years relative to HS dropouts.
The population the grid targets has not shared in average health gains. Grid-based awards overwhelmingly involve high-school graduates and dropouts doing physically demanding work — precisely the SES group whose health has improved least. Raising thresholds would deny benefits to workers who cannot realistically transition to other work, not to workers who can.
Labor market changes do not favor this population. The shift toward cognitively demanding work generally benefits older workers — but since the late 1990s, job opportunity gains for older workers have gone primarily to better-educated workers. The grid's rationale for less-educated applicants remains intact.
A literature review (Mann, Stapleton, and de Richemond 2014) found no empirical support for the specific cutoff ages, but concluded it was unrealistic to expect such precision — the grid sets administratively consistent bright-line rules, not actuarially calibrated thresholds.

Residual Work Capacity at the Age-55 Threshold (Strand and Messel 2019)

Strand and Messel (2019) provide the most direct causal test of the claim underlying grid reform proposals: that older workers near the age-55 threshold retain substantial work capacity that the grid's generous eligibility criteria suppress. Using the MMS (2013) examiner IV on Disability Operational Data Store, Detailed Earnings Record, and Case Processing and Management System (DIODS/DER/CPMS) administrative data, they estimate the work-disincentive effect of DI allowance separately for ages 55–56 (subject to the lenient age-55-and-over grid rules) and ages 52–53 (subject to the more stringent pre-55 rules).

Key IV estimates (above-Substantial Gainful Activity [SGA] employment):

Age group	IV estimate (above SGA)	Always-takers	Marginals (compliers)
Ages 55–56 (lenient grid)	−11.2 pp***	76.7%	17.3%
Ages 52–53 (stringent grid)	−12.9 pp***	—	—
Difference	−1.7 pp (n.s.)

First stage: coefficient 0.173; Angrist-Pischke F = 271.5.

The −11.2 pp estimate is an upper bound on residual work capacity for marginal applicants at the age-55 threshold — lower than MMS 2013's −16.6 pp for the full applicant pool, consistent with older marginal applicants being more severely impaired. The 1.7 pp difference across the threshold is not statistically significant: work discouragement is essentially the same whether applicants are adjudicated under lenient or stringent grid rules.

The pathway finding. The primary empirical effect of the age-55 cutoff is on the pathway to allowance, not on final allowance rates. Before age 55, the majority of vocational-grid allowances occur at the ALJ hearing level; after age 55, most occur at the initial DDS level. The cutoff makes it easier to receive an initial award, not materially more likely to receive any award eventually. Among denied applicants, 71.3% of ages 55–56 and 68.5% of ages 52–53 appeal or reapply — a 2.8 pp difference that is weakly significant.

Policy implication for Congressional Budget Office (CBO) 2012 projections. CBO estimated that raising the grid's age thresholds by 2 years would reduce awards by ~50,000 by 2022 and save $5.8 billion over 10 years. Strand and Messel's evidence implies these projected savings would primarily reflect slower pathways to allowance — more applicants cycling through appeals before receiving identical eventual outcomes — rather than sustained increases in labor force participation. The delay channel itself carries costs (foregone benefits during extended waits, the deterrence effects documented in Autor et al. 2015) that the CBO projection did not account for.

Method comparison. Applying Autor, Maestas, Mullen, and Strand (2015)'s delay-channel IV (which adds examiner average processing time as a second instrument for time out of the labor force) produces a larger estimate: −23.7 pp above SGA for ages 55–56. The delay-channel estimate is larger because it captures the behavioral effect of the application process itself on subsequent labor market attachment — making it the right counterfactual for a policy that lengthens the path to allowance (e.g., reduced initial-stage approval leading to more appeals), but not for a policy that simply changes the allowance standard while holding application incentives constant.

Policy Implication

Raising the vocational grid's age thresholds (e.g., from 50 to 55) would likely increase mortality among near-threshold claimants by denying benefits to individuals who qualify under current rules, rather than returning them to work. Strand and Messel (2019) provide causal evidence that age-threshold changes produce delays in the pathway to allowance rather than labor-market reentry: the work-disincentive effect of DI is statistically identical on both sides of the age-55 cutoff, and the majority of denied applicants appeal or reapply. The regressivity of such a change is compounded by the SES gradient in health improvement: low-income workers near threshold ages already face far worse mortality than national averages used in the "Americans are living longer" argument.

Maestas (2019) Reform Proposal: Replace Grid with WD-FAB/ORS System

Maestas (2019) offers the most systematic case for eliminating the grid, identifying three structural limitations:

Obsolete occupational database. Grid rules assume a national supply of "unskilled" jobs drawn from the Dictionary of Occupational Titles (DOT; final publication: 1990s). Industrial automation has restructured the physical demands of many surviving jobs, and advances in assistive technology have altered functional thresholds — but the grid has not been updated.
Single-domain RFC. Residual functional capacity is conceptualized only in terms of physical strength (sedentary/light/medium/heavy). Non-exertional limitations — mental impairments, sensory disorders, pain — receive only secondary treatment despite now accounting for the majority of new awards.
Coarse fixed age thresholds. The grid's binary disability decisions at ages 45, 50, and 55 ignore individual biological variability in aging and the well-documented SES gradient in health improvement (gains in healthy life expectancy have gone primarily to higher-SES workers, not the low-education manual labor demographic the grid serves).

Proposed replacement: WD-FAB linked to ORS.

Work Disability Functional Assessment Battery (WD-FAB): 8 physical and behavioral domains (Basic Mobility, Upper Body Function, Fine Motor Function, Community Mobility, Communication and Cognition, Mood and Emotions, Resilience and Sociability, Self-Regulation); 300+ hierarchically structured items; computerized adaptive testing; 15–20 minute administration; validated in SSDI applicant samples (Marfeo et al. 2013, 2015). Current limitation: items reference home-domain functioning, not work tasks — a crosswalk to job demands is needed.
Occupational Requirements Survey (ORS): SSA/Bureau of Labor Statistics (BLS) employer-side establishment survey designed to replace the DOT; nationally representative (~90% of civilian workers as of 2018); measures job requirements along dimensions that could be aligned with WD-FAB functional domains.
Process: Measure individual functional capacity with WD-FAB → query ORS database of job requirements → identify feasible occupations → assess skill transferability → rank by earnings → second-ranked occupation's earnings = estimated residual earnings capacity → percent earnings loss = disability rating.
Netherlands model as international precedent: a social security physician administers a functional assessment; an algorithm matches the applicant to a database of occupations; a labor expert prunes the list; residual earnings capacity is estimated from ranked feasible jobs; benefit amount is proportional to earnings loss.

This proposal is directly linked to Nicole Maestas's second reform — introducing partial disability benefits — which depends on credible, individualized earnings capacity estimates that the current grid cannot provide. See Maestas 2019 — Identifying Work Capacity and Promoting Work.

International Analogue: Austrian Two-Tier Eligibility Structure (Staubli 2011)

Austria operates a structurally similar age-based occupational standard that provides a rare international test of how such thresholds affect behavior. Austrian DI uses two tiers:

Below age 55: claimant must be unable to work in any reasonable occupation in the national economy ( $100+$ qualifying jobs nationally) — directly analogous to SSA's "any work" standard for applicants below the vocational grid's key age thresholds.
Above age 55: claimant need only be unable to work in a similar occupation to their past work — analogous to the grid's more lenient standard for older workers with limited transferable skills.

This two-tier cliff generates a sharp enrollment spike at age 55 in Austrian administrative data, just as SSA's grid generates visible mortality dips at ages 50 and 55.

Austria's 1996 Structural Adjustment Act raised the relaxed-eligibility threshold from $55 \to 57$ for men only (women's early retirement age of 55 was unaffected). The difference-in-differences (DiD) design — treated: men 55–56; comparison: men 49–54 — provides clean causal identification. Key results:

Outcome	Main estimate (excl. transition years)
DI enrollment	$-6.0$ to $-7.4$ pp (base $22.6\%$ )
Employment	$+1.6$ to $+3.4$ pp
Unemployment	$+3.5$ to $+3.9$ pp
Total labor force participation (LFP) (incl. unemployment insurance (UI)/sickness insurance (SI))	$+6.1$ to $+7.5$ pp

Collar heterogeneity confirms the mechanism: blue-collar workers (pre-reform DI base $33.3\%$ ) show DI $-8.9$ to $-11.2$ pp and employment $+2.7$ to $+6.0$ pp. White-collar workers (pre-reform DI base $7.5\%$ ) show DI $-1.3$ to $-1.6$ pp and employment null — because white-collar eligibility below age 55 was already assessed against the same occupational group, not "any occupation," so the grid's bite was less sharp.

Postponement, not deterrence: The transition analysis shows enrollment at ages 55–56 drops $\approx 10$ pp immediately post-reform but fully recovers at ages 58–59. Most workers simply waited for the new threshold rather than permanently exiting the DI pipeline — the Austrian analogue of the "denial roundabout" documented in French and Song (2014) for the US.

Spillovers dominate: DI, unemployment insurance, and sickness insurance act as close substitutes; the reform "succeeded only partially" because UI absorbed most of the DI enrollment reduction. Net employment gain ( $+1.6$ to $+3.4$ pp) is substantially smaller than the DI reduction ( $-6.0$ to $-7.4$ pp), with the gap absorbed by UI and SI take-up.

The Austrian pattern is consistent with the US vocational grid literature in one key respect: raising an age-based eligibility threshold shifts workers across insurance programs and time more than it returns them to sustained employment. See Stefan Staubli and Staubli 2011 — The Impact of Stricter Criteria for Disability Insurance on Labor Force Participation.

International Comparative Context (Rajnes & Notaro 2019)

The U.S. vocational grid is unusual among developed economies. A comparative survey of 11 Organisation for Economic Co-operation and Development (OECD) countries (US, Australia, Canada, Denmark, Finland, Ireland, Netherlands, Norway, Sweden, Switzerland, UK) finds:

Work experience is the only vocational factor (VF) present in all 11 countries — universally used to assess transferable skills and ability to perform suitable other work.

Age is an explicit determination factor in only a minority of countries. Finland and Norway treat age 55+ as suggesting that rehabilitation measures are less appropriate. Sweden uses age definitively to split programs (temporary benefit for ages 19–29; long-term for 30–64). Australia, Denmark, Ireland, Netherlands, Switzerland, and UK do not use age as an explicit eligibility factor. The U.S. age grid ( $\leq 49$ / $50$ – $54$ / $55+$ ) that drives the mortality fingerprint at ages 50 and 55 is thus internationally distinctive.

Education is explicitly used in determination almost nowhere besides the U.S. Denmark and the Netherlands consider it in labor market expert job-matching algorithms. Ireland, Switzerland, and the UK do not consider education at all. The U.S. four-tier education scale (illiteracy / marginal / limited / HS+) applied at Step 5 is the most prominent exception in the entire sample.

Partial disability is common but absent from the U.S. Incapacity thresholds: Denmark $50\%$ , Norway $50\%$ , Finland $60\%$ , Switzerland $70\%$ , Netherlands $80\%$ . The U.S. is the only surveyed country requiring full work incapacity for any benefit.

Sick-leave staging precedes long-term disability application in most countries (often $12+$ months of sickness benefit required first). The U.S. has no mandatory precursor sickness stage.

Employer involvement is mandatory in early stages in Finland, Netherlands, Norway, and Sweden. The Netherlands requires a joint employee-employer reintegration plan before disability application. This is absent from U.S. law.

Rehabilitation-first orientation: Denmark and Norway assess functional capacity primarily to determine rehabilitation appropriateness before benefit eligibility. The U.S. applies VFs only after medical severity has been established — a fundamentally different theory of what determination is for.

ICF adoption: Only Sweden has compulsorily integrated the World Health Organization (WHO)'s International Classification of Functioning (ICF, 2001) categories into its determination process despite universal endorsement.

The grid reform debate (2005 NPRM, 2009 withdrawal, 2015 ANPRM) proposes raising U.S. age thresholds to match a claimed general-population health improvement. The international comparison shows that no peer country uses hard age thresholds the way the U.S. does; the reform debate is thus a distinctly American policy question with no direct international comparator to test against. See David Rajnes and Tony Notaro.

Age Discontinuity as a Cyclicality Instrument (Carey, Miller, and Molitor 2025)

The age-50/55 entry spikes documented in DI mortality studies — previously treated primarily as a policy design feature — turn out to be a powerful instrument for identifying the mechanism behind DI's countercyclical behavior.

Carey, Miller, and Molitor (2025) use Medicare administrative data ( $15.79$ million DI beneficiaries, 1993–2017) and exploit the age-50 grid discontinuity to separate health shocks from entry costs in recession-driven DI entry. The logic: the age-50 threshold admits additional entrants via pure entry-cost reduction (lower eligibility hurdle) without any direct effect on health. This makes it an instrument for the entry-cost channel, allowing the authors to trace out what the health of marginal entrants would look like if driven only by lower entry costs.

Key findings:

Entry rate spikes $67\%$ at age 50 ( $382 \to 636$ per million/month) and similarly at 55
Entry sensitivity to unemployment jumps $2$ – $4\times$ above age 50: 49-year-olds add $7.8$ per million per 1 pp unemployment rate (UR); 50-year-olds add $41.8$ — the flatter eligibility cost curve for 50-year-olds means any given decline in expected earnings activates far more entrants
Workers 50+ generate two-thirds of all recession-induced DI entry despite comprising half of all awards
Using the Marginal Treatment Effects framework (Heckman-Vytlacil 2005), the marginal spending functions at mean and high unemployment are statistically indistinguishable $\to$ health shocks explain only $1.5\%$ of DI cyclicality (confidence interval (CI): $0.1\%$ – $3.2\%$ )
Welfare: workers 50+ receive $\approx 2\times$ the insurance value from DI's countercyclical protection relative to younger workers, because their larger entry response magnifies the aggregate surplus from the benefit-cost gap

This use of the grid discontinuity as an instrument for entry-cost variation is methodologically complementary to Chen and van der Klaauw (2008)'s RD identification of work-disincentive effects: both exploit the same institutional discontinuity but ask different questions (work-disincentive vs. cyclicality mechanism).

See DI Countercyclicality for the full mechanism and Carey Miller and Molitor 2025 — Why Does Disability Increase During Recessions Evidence from Medicare.

Open Questions

What share of total DI mortality improvement since 1978 is attributable to the compositional effect of vocational awards?
The 1970s subperiods show a similar (if smoother) flattening of mortality at older ages, suggesting informal age considerations predate the formal 1978 grid.