Summary
Extended and more methodologically rigorous follow-up to Friedman et al. (2016), using 44.9 million young adults (Internal Revenue Service (IRS) admin data 2002–2015, birth cohorts 1980–1993). Key advances: separates sorting from causal place effects using signal correlations from Chetty-Hendren (2018b) mover estimates; consolidates local characteristics into eight confirmatory factor analysis (CFA) factors; documents cyclicality concentrated among poor children; and directly compares disability insurance (DI) take-up to income mobility across commuting zones (CZs).
Key Claims
- Cyclicality concentrated at the bottom: DI first-entry hazard co-moves with unemployment nationally, but the absolute variation is roughly 3× larger for bottom income quintile children vs. top (standard deviation [SD] of hazard rate: 0.036 percentage points [pp] vs. 0.012 pp; from 2006–2010 Great Recession trough to peak, bottom quintile entry rose 0.060 pp vs. 0.020 pp for top quintile). When age is held fixed, unemployment explains 52–53% of the variation in entry hazards in both quintiles, but the absolute swings are far larger at the bottom.
- Geographic variation robust: Same finding as 2016 paper at both state and CZ level. Maine, New Hampshire, and Vermont have >5% of poor children ever on DI. Nevada and Utah are among the lowest. Rich children show virtually no state-level variation (nearly all states <1.3% for top quintile).
- Sorting vs. causal decomposition: Signal correlations from mover estimates reveal which local characteristics are causal vs. driven by sorting. Sorting-driven (disappear in movers): racial segregation, income inequality, mean household income, migration rates, manufacturing share. Causal (persist in movers): education quality (association actually strengthens), school spending per pupil (observationally near zero, positive in movers), taxes (lower local tax burden → higher DI persists). Attenuated but persistent: social capital (largest single correlate observationally, roughly halved for movers).
- DI vs. income mobility: null overall: Correlation between CZ-level DI take-up and absolute income mobility (Chetty-Hendren 2018b) is −0.22 for mover estimates — effectively null to weakly negative. Both outcomes correlate with the same local characteristics in the same direction (education, social capital → both higher DI and higher mobility). Exception: migration and family structure matter for mobility but not DI.
- "Good CZs" are rural: Places that simultaneously produce low DI and high income mobility (top quartile mobility, bottom quartile DI) are unusually small — average population 128K vs. 521K for median CZs. Low DI in rural areas may partly reflect limited access rather than low underlying disability.
- Social Security Administration (SSA) field office access: CZ-level presence of at least one SSA office is modestly and positively correlated with DI take-up observationally, consistent with lower application costs. Mover estimates suggest the observational correlation is substantially driven by sorting.
- Factor structure (CFA): Eight latent factors (segregation, inequality, taxes, education, local labor, migration, social capital, family structure) account for the local characteristic correlations; taxes and education most robustly predict DI take-up in mover estimates.
Concepts Introduced or Extended
Entities Mentioned
Quotes
"The places that generate high DI take-up among these children tend to be places which have otherwise 'good' observable characteristics: low segregation, better schools, more income, and less income inequality."
"The overall relationship between DI take-up and income mobility across CZs is in aggregate null, with some evidence that places performing well on both measures are unusually less populous."
My Take
This paper significantly advances the 2016 conference version by disentangling sorting from causal mechanisms and providing a richer characterization of local factors. The key methodological contribution is applying the Chetty-Hendren (2018b) signal-correlation framework to DI, which is a non-trivial adaptation. The "good places paradox" persists in causal estimates for education and taxes, lending it more credibility than the 2016 cross-sectional version. The null DI–income mobility relationship is surprising and under-explored; the paper correctly identifies it as a puzzle for future research. The rural "good CZs" finding raises the important identification concern that low DI in those areas may reflect limited access, not low disability rates.
Sources