DI Beneficiary Mortality

Definition

The mortality experience of Social Security Disability Insurance (DI) disabled-worker beneficiaries, measured as period death probabilities by age, sex, and duration of entitlement (years on the rolls). DI beneficiary mortality systematically and dramatically exceeds general-population mortality at every age and every duration studied, from the 1970s through 2015.

Key Ideas

Duration effect: Mortality is highest in year 0 (first year of entitlement) and falls with each additional year survived on the rolls. After 10+ years, beneficiary mortality is much lower than at year 0 but still far exceeds the general population.
The 1918 benchmark: DI beneficiaries aged 25+ in their first year on the rolls face higher mortality than the general public did during the 1918 influenza pandemic — in every study period through 2015.
Gender-age paradox: In absolute terms, older and male beneficiaries have higher mortality. In relative terms (ratio to general population), female and young (25–35) beneficiaries are worse off. Young female beneficiaries faced ratios up to $70\times$ the general population in 1977–1980.
Mortality improvement: Beneficiary mortality has improved faster than the general population's, especially at year 0 and younger ages, but the gap remains large.
Within-5-years mortality: 1 in 5 male beneficiaries (~~20%) and ~1 in 6 female beneficiaries (~~17%) die within 5 years of entitlement (Zayatz 2011, Actuarial Study No. 122). This figure, widely cited by the Social Security Administration (SSA) in program defense, is consistent with the select-period mortality ratios: given 8.5–9.9 $\times$ excess mortality at year 0 for a 50-year-old, a ~20% 5-year cumulative death rate follows for a mixed-age, recently-awarded cohort concentrated in poor health.

How It Works

SSA's Office of the Chief Actuary (OACT) constructs actuarial period mortality tables using a "select and ultimate" structure. For each age at entitlement, mortality is tracked separately for the first 10 years on the rolls (the "select period"), after which duration no longer independently predicts mortality and only attained age matters (the "ultimate" period).

The select structure captures a key biological/statistical fact: the probability of death conditional on having just qualified for DI is much higher than the probability conditional on having survived on the rolls for years. New entrants are acutely ill; survivors have already demonstrated relative resilience.

Death probabilities are calculated in a multiple-decrement environment — DI exits include death, recovery, and conversion to retirement benefits — so the mortality tables are conditioned on still being an active DI beneficiary.

Why It Matters

Policy: Life expectancy at age 50 for a DI beneficiary in 2011–2015 was 9–10 years shorter than for a 50-year-old in the general population. This affects program design, Continuing Disability Review (CDR) policy, and vocational grid thresholds.
Program integrity: Mortality statistics are confounded by who enters the rolls (see Disability Insurance Chronology). Declining beneficiary mortality can reflect improved health or a compositional shift toward less-deadly diagnoses.
Causal effect of benefits: Alexander Gelber, Timothy Moore, and Alexander Strand (2017) established using a Regression Kink Design that DI income causally reduces mortality — benefits themselves save lives, particularly for lowest-income beneficiaries. See below.

Causal Effect of DI Income on Mortality (Gelber, Moore, and Strand 2017)

Identification

The DI benefit formula converts Average Indexed Monthly Earnings (AIME) to a monthly Primary Insurance Amount (PIA) through a piecewise linear function with three "bend points" where the marginal replacement rate changes discontinuously. Medicare eligibility and all other program rules are smooth across these points — only benefit size changes. A Regression Kink Design (RKD) uses the slope change in the benefit schedule to identify the causal effect of benefit size on mortality, separately from any effect of DI eligibility.

Results (3.65 million new DI beneficiaries, 1997–2009; outcome = annual mortality rate, first 4 years)

Bend point	AIME percentile	Mean annual DI	Effect of +$1,000/yr	Elasticity
Lower	4th	$8,543	−0.26 percentage points (pp)	−0.56
Family maximum	30th	$12,648 +$ 6,324 (dependent)	−0.09 pp	−0.57
Upper	84th	$20,777	No robust effect	—

Effects concentrate at the bottom: the income-mortality relationship is concave, and the marginal dollar does the most to extend life for those with the least.

Cost-Effectiveness

Bend point	Cost per statistical life-year saved
Lower	$58,574 ($ p < 0.05$)
Family maximum	$236,626 ($ p < 0.05$)

The Value of a Statistical Life Year (VSLY) lower boundary recommended by expert panels is $50,000. At the lowest-income group, DI income passes the cost-effectiveness threshold. These mortality gains were entirely absent from prior welfare analysis of the program.

Mechanism (Suggestive)

The mortality effect is not mediated by labor supply — earnings show no response to DI income at the lower or family maximum bend points. Survey evidence (Consumer Expenditure Survey [CES], Survey of Income and Program Participation [SIPP]) shows that DI households' income-to-expenditure elasticity is 52% higher than non-DI households, concentrated in food, housing, utilities, healthcare, and transportation. The 24-month Medicare waiting period makes early receipt particularly consequential. Largest effects are for cardiovascular conditions and cancer — health-care-amenable conditions where additional income directly enables life-extending treatment and nutrition.

10-Year Mortality Among New Awardees (Anand and Ben-Shalom 2016)

In a 2001 award cohort followed 10 years via administrative records, 21.3% of DI-first awardees and 20.7% of Supplemental Security Income (SSI)-first awardees died within the 10-year window. Death was the second most common single milestone for DI-first awardees (after no-milestone at 32.5%) and exceeded Full Retirement Age (FRA) attainment for SSI-first awardees (20.7% vs. 14.7%), reflecting SSI's younger and more disadvantaged population. These figures are consistent with the OACT period mortality tables: given that DI beneficiaries have year-0 mortality ratios of 8.5–9.9 $\times$ the general population at age 50, a 21% cumulative death rate within 10 years of award is expected for a mixed-age cohort concentrated in poor health.

Key Quantitative Benchmarks (2011–2015)

Age	Metric	Men	Women
50	Mortality ratio (year 0 vs. general pop.)	$8.5\times$	$9.9\times$
50	Life expectancy gap (year 0)	−10.1 yrs	−9.7 yrs
50	Mortality ratio (10+ years on rolls)	$3.9\times$	$4.5\times$
35	Mortality ratio (year 0)	$\approx 18.5\times$	$\approx 30\times$

Relationship to the Income-Mortality Gradient

Chetty et al. (2016) found that men in the bottom 1% of the U.S. income distribution have a life expectancy of 72.7 years at age 40 — 14.6 years below the top 1%. DI beneficiaries are overwhelmingly low-income and cluster at or below this threshold. This means the DI/general-population mortality gap documented by Meseguer (2021) and the income-LE gradient documented by Chetty et al. are measuring overlapping phenomena: much of what appears as "DI excess mortality" relative to the general population reflects the income-LE gradient, since the comparison denominator (general population) averages across all income levels while the DI numerator is concentrated at the bottom.

The geographic pattern reinforces this: the commuting zones with the worst life expectancy for low-income individuals — Gary IN, Detroit MI, Toledo OH, Oklahoma — are the same regions with the highest DI application rates. See Income-Mortality Gradient.

DI Exit Probabilities by Cause: Competing-Risks Framework (Raut 2017)

Conventional Kaplan-Meier survival analysis treats competing exit causes (death, recovery, Old-Age and Survivors Insurance (OASI) conversion) as independent — a biased assumption when three exits are simultaneous. Raut (2017) applies the Fine-Gray subdistribution hazard model to Continuous Work History Sample (CWHS) data for 1981–2000 entrants to estimate cumulative incidence functions for each cause jointly.

30-Year Overall Exit Probabilities (All Impairments, All Ages)

Exit cause	30-year cumulative probability
Death	$\approx 38\%$
Recovery	$\approx 8\%$
OASI conversion	$\approx 39\%$
Still on rolls	$\approx 15\%$

Death is the modal exit cause across the full beneficiary lifecycle. Recovery — at only 8% over 30 years — is strikingly rare for the overall population.

Exit Probabilities by Age at Entitlement (9-Year Window)

Age at entitlement	Recovery	Death
20–30	16.3%	12.9%
31–40	8.3%	21.0%
41–50	3.8%	29.8%
51–55	1.4%	32.1%

Young entrants are recovery-dominant; older entrants are death-dominant. The gradient is steep: a beneficiary entering at ages 20–30 is $11\times$ more likely to recover within 9 years than one entering at ages 51–55.

Exit Probabilities by Impairment Type (9-Year Window, Ages 20–30 and 51–55)

Impairment	Age	Recovery	Death
Musculoskeletal	20–30	32.4%	4.4%
Mental disorders	20–30	22.1%	2.8%
Circulatory	20–30	15.6%	29.5%
Neoplasms	20–30	15.1%	69.8%
Neoplasms	51–55	1.4%	89.6%

This table reveals the core mechanism linking the compositional shift to declining beneficiary mortality: musculoskeletal and mental conditions have the lowest death rates AND the highest recovery rates among all impairment types. Yet even their recovery rates (32.4% for young musculoskeletal entrants over 9 years) leave the majority of entrants on the rolls long-term. Neoplasms are overwhelmingly death-dominant (70–90% cumulative death within 9 years at all ages).

Sex Asymmetry in Exit Probabilities

Among young (ages 20–30) entrants with musculoskeletal disorders: men's 9-year recovery probability is 38.5% vs. women's 20.3%. This pattern repeats across impairment types — women have lower death exit rates AND lower recovery exit rates than men of the same age and impairment. The double-low structure means women remain on the rolls for longer average durations, contributing to the rising female share of DI enrollment. See DI Growth Decomposition.

1980s vs. 1990s Cohort Comparison

Death exit rates declined for all subgroups between the 1981–1985 and 1991–1995 entrant cohorts. Recovery exit rates were largely stagnant — with modest improvement only for ages 41–50. This is the key empirical result connecting DI mortality to the Morbidity-Mortality Distinction: declining mortality on the rolls reflects medical advances and compositional shifts toward lower-mortality conditions, not improved work capacity.

Work Disability as Independent Mortality Predictor (Heiss et al. 2007)

Heiss, Börsch-Supan, Hurd, and Wise (2007) estimate a latent-health model on all four Health and Retirement Study (HRS) cohorts ( $n \approx 25{,}000$ ; waves 1–6, 1992–2002) that jointly addresses selective mortality, state dependence, and misclassification error in observed health reports.

Key results for DI beneficiary mortality:

Work disability → mortality (odds ratio [OR] = 1.76): Even after controlling for self-rated health (SRH), work disability independently predicts death. The two health dimensions carry separate mortality information — SRH is not a sufficient proxy for functional limitation.
Survival gap: Conditional on being alive at age 50, survival to age 70 is 91.6% for those in excellent health vs. 48.5% for those in poor health — a 43 percentage-point gap.
Selective mortality bias: Observed health transitions from Markov models dramatically underpredict the persistence of poor health over 8 years. Individuals in poor health are most likely to die and exit the panel, so the surviving sample appears healthier than the true trajectory of the initial cohort. The Heiss et al. latent-health autocorrelation ( $\rho \approx 0.98$ for SRH, $\rho \approx 0.97$ for work disability, 1-year; 0.7/0.5 at 20 years) captures this persistence.

The selective mortality finding has a direct implication for interpreting DI beneficiary mortality trends: if DI rolls are dominated by individuals in poor health (which they are), then any longitudinal analysis of the rolls' "average" health will be confounded by the exit of the sickest members via death — making the surviving beneficiary population appear to improve faster than the population of initial entrants actually does.

Causal Effect of DI Allowance on Mortality (Black, French, McCauley, and Song 2024)

Identification and Estimand

Black et al. (2024) use the same ALJ judge leniency IV as French and Song (2014) — leave-one-out allowance rate within hearing office-day pairs — but replace the labor supply outcome with 10-year mortality. This isolates the causal effect of allowance (receiving vs. being denied benefits), as distinct from Gelber/Moore/Strand (2017) who identify the effect of benefit amount for existing recipients. The two designs answer complementary questions: how much does the dollar size of benefits matter (RKD) vs. does getting onto the rolls at all change mortality (judge IV)?

Sample: 610,231 ALJ-stage appellants aged 55–64, 1995–2004; 1,436 judges.

LATE and MTE Heterogeneity

LATE: +2.8 pp 10-year mortality for marginal appellants. A naive reading: DI allowance increases mortality for the complier population.
MTE curve: The Local Average Treatment Effect (LATE) is the integral of the Marginal Treatment Effect (MTE) over the range of propensity scores induced by the instrument. The MTE curve is downward-sloping in the latent health index:
- Inframarginal recipients (very sick; allowed by any judge) → allowance reduces mortality. The cash income and Medicare/Medicaid channel (~$47k discounted to age 65) enables life-extending healthcare.
- Marginal recipients (relatively healthy; allowed only by lenient judges) → allowance increases mortality. The work-disincentive channel (−8 pp employment at 5 years) removes the survival benefits of employment.
The positive LATE is driven by the composition of compliers: lenient judges allow the healthier tail, so the LATE estimates the effect precisely where work disincentives dominate.

Condition Heterogeneity

Condition	Direction of effect	Mechanism
Cancer, respiratory disease	Negative / near-zero	Insurance channel dominates; high mortality, high healthcare costs
Musculoskeletal	Positive	Work-disincentive channel dominates; lower mortality, strong labor force participation (LFP) response

This mirrors the French/Song (2014) subgroup ordering: musculoskeletal has the largest LFP response (−28.5 pp at 3 yr), so it is precisely the condition where DI allowance most disrupts employment and generates the worst mortality trade-off.

Three Channels

Cash income + Medicare/Medicaid: ~$47k in present-value transfers (discounted to age 65) — reduces mortality via healthcare access, nutrition, housing stability
Work disincentives: −8 pp employment at 5 years — increases mortality by removing occupational health benefits and social engagement
Medicaid/SSI effects for lowest-income applicants: additional insurance pathway concentrated at the bottom of the income distribution

Policy Implication

Current ALJ thresholds are approximately correctly calibrated given the evidence: the marginal applicant's mortality cost from allowance is real but modest, and the inframarginal population benefits substantially. Condition-specific targeting (stricter thresholds for musculoskeletal; more lenient for cancer/respiratory) could reduce mortality costs without large welfare losses.

Open Questions

To what extent does mortality improvement reflect genuine health gains vs. compositional shifts in the beneficiary pool? Liebman (2015) provides the strongest quantitative case for composition: the shift from circulatory/cancer to musculoskeletal/mental conditions is large enough to account for "a significant portion" of the decline. The 21% enrollment counterfactual (see DI Growth Decomposition) implies the condition-mix change is substantial, and lower-mortality conditions mechanically reduce the measured death rate.
No actuarial data exists for the 1980s — a significant gap given the CDR crisis of that era.
How will the rise in "deaths of despair" in the general population affect the DI/general-population mortality gap post-2015?
How much of the DI/general-population mortality gap is explained by the income-LE gradient alone, independent of health status?
What is the mortality rate of the DI Denied Population? Weaver (2020) shows denied applicants have health profiles nearly as severe as approved beneficiaries, suggesting their mortality substantially exceeds the general population's — but no systematic data exists.