Geographic Variation in Disability Insurance

Definition

The empirical phenomenon that Disability Insurance (DI) receipt rates among young adults vary enormously across US commuting zones (CZ) and states, but this variation is almost entirely concentrated among children from low-income families. Rich children's DI rates are near-identical across places; poor children's rates can differ by a factor of $3$ – $5$ depending on where they grew up. Established by Friedman, Lurie, Mogstad, and Chetty (2016) and extended by Friedman, Kellogg, Lurie, and Mogstad (2018) using Internal Revenue Service (IRS) administrative tax data linking birth cohorts to parental income.

Key Ideas

Intergenerational income gradient: DI receipt is strongly and continuously negatively correlated with parental income. At ages $24$ – $34$ , the net DI hazard rate is $\approx 4.8\times$ higher for children from the bottom $1\%$ of the parental income distribution than the top $1\%$ ( $20.1$ vs. $4.2$ per $10{,}000$ ; bottom quintile $1.1\%$ vs. top quintile $0.3\%$ at age 24). The gradient is stable across ages and holds conditional on parents not receiving DI themselves.
Geographic variation concentrated at the bottom: Virtually all cross-area variation in DI rates occurs among poor children. Rich children's DI rates are near-identical regardless of where they grew up. This is why simple geographic comparisons conflate income composition with place effects.
Causal neighborhood effects ( $\approx 50\%$ ): Using the Chetty-Hendren (2018) movers design — comparing siblings who moved at different ages — roughly $50\%$ of the cross-CZ variation in young adult DI rates is causal. The exposure slope is $-0.032$ per year of childhood spent in a place (significant through age 22, near zero after), confirming formative childhood years matter. This is parallel to the income mobility literature.
Cyclicality concentrated at the bottom: DI first-entry hazard is more cyclical for poor children. From 2006 to 2010, the bottom-quintile entry hazard rose $0.060$ pp (from $0.17\%$ to $0.23\%$ ) while the top-quintile rose only $0.020$ pp — roughly a $3:1$ ratio in absolute swings.
Sorting vs. causal mechanisms (Friedman et al. 2018): Among local characteristics, education quality (student-teacher ratios, test scores, school spending) and tax burden (lower taxes $\to$ higher DI) appear genuinely causal — their correlations with DI persist or strengthen in mover estimates. Segregation and income inequality are largely sorting-driven; they vanish in mover estimates. Social capital (the single strongest observational correlate) is attenuated roughly by half in mover estimates but remains positive — partially causal, partially sorting.

The "Good Places" Paradox

The most striking finding is that the CZs with the highest DI rates for poor children tend to score well on most standard measures of place quality: lower income inequality, lower income and racial segregation, better schools, and higher social capital. These are also the characteristics that Chetty et al. (2014) and Chetty and Hendren (2018) associate with high income mobility for poor children.

This contradicts the intuitive narrative that DI receipt is driven by economic distress. The "bad" labor-market indicators — high manufacturing share, high Chinese import exposure, high teen labor force participation — are also positively correlated with high-DI CZs, but this relationship largely disappears in mover estimates (sorting). Local unemployment rates show no cross-sectional correlation with DI despite DI entry being strongly pro-cyclical nationally — a striking disconnect between cross-sectional and time-series evidence.

DI Take-Up vs. Income Mobility

The overall relationship between DI take-up and income mobility across CZs is null to weakly negative (mover-estimate correlation: $-0.22$ ). Both outcomes share many of the same correlates (education, social capital, taxes $\to$ higher DI and higher mobility), yet the direct cross-CZ relationship is near zero. The places that produce both low DI and high income mobility (top quartile mobility, bottom quartile DI) are unusually rural — average population $128\text{K}$ vs. $521\text{K}$ for the overall CZ distribution — which raises the possibility that low DI in those areas reflects limited access rather than low underlying disability rates.

Variance Decomposition: Prevalence vs. Participation vs. Administration (Gettens et al. 2018)

Gettens, Lei, and Henry (2018) are the first to use variance decomposition — rather than regression — to analyze geographic variation in DI/Supplemental Security Income (SSI) participation. The key insight is that the overall participation rate is the product of two separate processes: disability prevalence (how common disability is in an area) and program participation conditional on disability (what share of people with disabilities actually receive DI/SSI). These have different determinants and should be decomposed separately; prior regression-based studies conflated them.

Method: CAPUMAs — 937 county-aligned substate units constructed by intersecting 2,069 American Community Survey (ACS) Public Use Microdata Areas (PUMAs) with 3,142 Social Security Administration (SSA) counties — using 2009–2011 data. Principal-components analysis further decomposes participation-among-disabled into uncorrelated socioeconomic subcomponents.

Main finding: ~90% of geographic variation in DI/SSI participation is explained by disability prevalence, socioeconomic characteristics, and their correlation. Administrative inconsistency is not a major driver — directly refuting Social Security Advisory Board (2001, 2012a, 2012b) concerns.

DI decomposition:

67.9% of variance → variation in disability prevalence
21.2% → variation in DI participation among disabled persons
10.9% → correlation between the two
Of the 21.2%: unobserved state-level factors (policy, administration) explain at most ~7 pp — the empirical upper bound on administrative inconsistency

SSI decomposition (different story — means-tested):

37.5% disability prevalence, 36.7% participation among disabled, 25.8% correlation
SSI participation-among-disabled driven by economic disadvantage: Black/low self-employment subcomponent (18.3%), public assistance participation (10.3%), income (8.1%)

DI-specific finding: Areas with more Hispanic residents, non-English speakers, foreign-born, and non-citizens have systematically lower DI participation (Hispanic/non-English subcomponent: 10.4% of participation-among-disabled variance). Probable mechanisms: insufficient work quarters for insured-status eligibility, language barriers, immigration status concerns.

Geographic patterns: DI ranges 1.0%–16.6% at CAPUMA level. Highest in Appalachian CAPUMAs (eastern KY, WV, TN, AL, AR, MS); lowest in major urban areas (Chicago 3.2%, Houston 2.6%, NE corridor) and Great Plains/Mountain West states.

Relationship to Friedman et al.: Gettens et al. operate at the full working-age population level with a descriptive variance decomposition; Friedman et al. use cohort-based movers designs to estimate causal neighborhood effects on young-adult DI rates. Both conclude variation is real and structural, not administrative. A key open question from Gettens — what explains geographic variation in disability prevalence — is related to but distinct from Friedman's question about what childhood environments causally affect DI take-up.

State-Level Application Rate Variation (Coe et al. 2011)

While Friedman et al. examine receipt rates among young adults using tax data, Coe, Haverstick, Munnell, and Webb (2011) study Social Security Disability Insurance (SSDI) application rates at the state level using Behavioral Risk Factor Surveillance System (BRFSS) + Current Population Survey (CPS) + Bureau of Labor Statistics (BLS) data across $862$ state-year observations (1993–2009). Mean state application rate is $0.83\%$ , ranging from $0.49\%$ (Utah) to $1.65\%$ (Mississippi).

Key findings:

Health, demographics, and employment explain $>70\%$ of cross-state variation ( $R^2 = 0.796$ without state fixed effects). Labor force participation is more predictive than the unemployment rate cross-sectionally.
Temporary disability insurance (TDI) mandates reduce SSDI applications: states with mandatory TDI programs have $0.117$ – $0.126$ pp lower application rates ( $p < 0.01$ ), concentrated in SSDI–SSI concurrent applicants (income-poor workers substituting TDI for SSDI).
Strict health insurance (HI) market regulation reduces within-state applications: guaranteed issue + community rating in the non-group market $\to$ $0.054$ pp lower applications ( $p < 0.01$ ), consistent with Employment Lock: accessible insurance outside employment reduces the value of SSDI as a Medicare pathway.
Republican governor effect: $-0.021$ pp within-state ( $p < 0.05$ ), concentrated in concurrent SSDI-SSI applicants; mechanism unspecified but plausibly welfare program administration and stigma.
SSDI-only vs. SSDI-SSI concurrent respond to different state factors: poverty and bad health predict concurrent; white race and smoking predict SSDI-only; the two populations are structurally distinct.

Why It Matters

Attribution problem: Geographic variation in DI is commonly attributed to economic distress and labor market weakness (Ruffing 2015). Friedman et al. show this is partially a sorting artifact; the causal drivers are different from the observational correlates.
Policy design: If neighborhood effects on DI are causal, interventions targeted at childhood environment (education, social capital, economic integration) might affect DI trajectories — but the "good places have more DI" finding complicates this inference.
Access vs. disability: The field-office access finding (observationally, SSA presence correlates with higher DI) and the rural "good CZ" puzzle both suggest that access barriers matter and that low DI take-up is not straightforwardly synonymous with low disability prevalence.

Open Questions

What are the mechanisms by which childhood place causally affects DI receipt? (Friedman et al. explicitly leave this open.)
Are similar place effects present at older ages?
Does the "good places" paradox reflect high demand for DI in well-functioning communities (awareness of benefits, lower stigma, more social capital enabling navigation of the application process) or genuine differences in health production?