Gettens et al 2018 — Accounting for Geographic Variation in Social Security Disability Program Participation

Summary

Gettens, Lei, and Henry (2018) decompose the wide geographic variation in Social Security Disability Insurance (DI) and Supplemental Security Income (SSI) participation rates across the United States into component factors using a novel variance decomposition approach applied to 937 county-aligned substate areas (CAPUMAs). Approximately 90% of the geographic variation in DI/SSI participation is explained by variation in disability prevalence and area socioeconomic characteristics — directly contradicting the assumption that administrative inconsistency is a major driver. For DI specifically, variation in disability prevalence alone accounts for 67.9% of total participation variance.

Key Claims

• ~90% explained: Geographic variation in DI/SSI participation is nearly fully accounted for by disability prevalence, socioeconomic characteristics, and their correlation — administrative inconsistency is not a major driver.
• DI variance decomposition: 67.9% attributable to variation in disability prevalence; 21.2% to variation in DI participation among persons with disabilities; 10.9% to correlation between the two.
• SSI variance decomposition: 37.5% disability prevalence, 36.7% participation among disabled, 25.8% correlation — unlike DI, SSI variation is roughly equally split between the two components.
• State policy upper bound: Fixed-effects regression estimates an upper limit of ~7 percentage points (pp) of the 21.2% DI participation-among-disabled variance as attributable to unobserved state-level factors, placing an empirical ceiling on what program administration can explain.
• DI demographics: The Hispanic/non-English subcomponent (10.4% of DI participation-among-disabled variance) is associated with lower DI participation — likely reflecting insufficient work quarters for insured status, language barriers, immigration status concerns, or different demand for DI.
• SSI economic disadvantage: The Black/low self-employment (18.3%) and public assistance participation (10.3%) subcomponents dominate SSI variation, reflecting SSI's means-tested structure and its link to concentrated economic disadvantage.
• Geographic patterns: DI participation ranges 1.0%–16.6% of working-age population at CAPUMA level. High in Appalachia (WV, KY, TN, AL, AR, MS); low in major urban areas (Chicago, Houston, Northeast corridor) and Great Plains/Mountain West.
• Justification bias small: American Community Survey (ACS) disability prevalence strongly correlates ($r$ = 0.8–0.9) with objective health indicators (mortality, diabetes, poor health, smoking), suggesting self-report bias does not substantially distort variance decomposition estimates.
• Methodological advantage over regression: Variance decomposition separates two distinct data-generating processes — what drives disability prevalence vs. what drives program participation conditional on disability — unlike regression-based $R^2$ decompositions (Rupp 2012, Coe et al. 2011, Ruffing 2015) that conflate them.
• CAPUMA geography: Novel substate unit combining 2,069 ACS Public Use Microdata Areas (PUMAs) and 3,142 counties into 937 county-aligned areas, enabling substate variance analysis unavailable in prior state-level research.

Concepts Introduced or Extended

• Geographic Variation in Disability Insurance
• Disability Insurance Program
• Substantial Gainful Activity

Entities Mentioned

• Social Security Administration
• John Gettens
• Alexander Strand

Quotes

"Our findings strongly suggest that geographic variation in program participation is mainly an indication of geographic variation in disability prevalence and socioeconomic characteristics and that inconsistency in program administration is not a major reason for the variation."

"If disability prevalence did not vary across areas, the geographic variation in DI participation would be reduced by approximately 80 percent and the variation in SSI participation would be reduced by approximately 63 percent."

My Take

The central methodological contribution — factoring total participation into disability prevalence × participation-among-disabled — is genuinely clarifying. Prior regression-based $R^2$ decompositions (Rupp 2012 estimated ~75%, Coe et al. 2011 estimated >70%, Ruffing 2015 estimated 84%) all produced comparable numbers from different variable sets, which the authors correctly interpret as evidence of high correlations between observed and unobserved factors — not convergent identification. The variance decomposition is more transparent about what it is: a descriptive accounting, not a causal claim.

The paper's exoneration of administrative inconsistency is substantively important given the Social Security Advisory Board's repeated concerns (2001, 2012a, 2012b). The ~7% upper bound for unobserved state policy effects is small relative to the total variation.

The main gap is that disability prevalence variation — the dominant component for DI — is left unexplained; the paper explicitly defers this to future research. Whether Appalachian disability prevalence reflects genuine incidence (occupational injury, substance use, poor healthcare) or differential out-migration of healthy workers is not resolved here and is arguably the more important policy question. The Hispanic/non-English finding is also underexplored: low DI among immigrants plausibly reflects insured-status coverage gaps (insufficient work quarters) rather than lower underlying disability, which has implications for extending coverage.