Conti Berndt and Frank 2009 — Early Retirement and DI-SSI Applications Exploring the Impact of Depression

Summary

Uses Health and Retirement Study (HRS) panel data (workers born 1931–1941, waves 2–5: 1994–2000) to estimate the direct and indirect effects of depression on early retirement and Disability Insurance (DI) / Supplemental Security Income (SSI) applications among later-middle-age workers. Depression alone raises DI/SSI application probability by $\approx 16\%$ for men — roughly equivalent to incident physical illness — and interacts multiplicatively with both physical illness and widowhood to further raise application rates. The interaction of depression and widowhood approximately fourfold increases the application probability over widowhood alone, especially for men's retirement outcomes.

Key Claims

• Mental illness is the fastest-growing cause of new DI/SSI claims since the early 1990s, concentrated in ages 45–64; major depression accounts for $\approx 30\%$ of all DI/SSI claims (Social Security Administration (SSA) 2004)
• Direct effect of depression: both estimation strategies confirm depression raises DI/SSI applications and early retirement and reduces employment for men and women; magnitude similar to incident physical illness
• Depression × physical illness (indirect): raises DI/SSI applications and early retirement above either alone; adding back pain to the illness definition eliminates the gender differential (men report more incident back pain)
• Depression × widowhood (indirect): widowhood alone has small impact on DI/SSI applications; interacted with baseline depression, raises application probability $\approx 4\times$ (Strategy 1); for men, depression × widowhood is the largest single driver of early retirement — consistent with the widower mortality effect (Christakis and Iwashyna 2003) having a labor-market analog
• Quantitative (Strategy 2, generalized estimating equations (GEE)): incident depression → $+16\%$ DI/SSI application probability (men), $+\approx 16\%$ (women); physical illness alone → $+25\%$ (men); depression × widowhood → $\approx 32\%$ for men ($\approx 2\times$ either alone)
• Center for Epidemiologic Studies Depression (CES-D) scale $\geq 4$/$8$ cutoff captures $\approx 79\%$ of Composite International Diagnostic Interview (CIDI) DSM-III major depression cases; CES-D overidentifies transient episodes → estimates likely downward-biased
• Prior economics literature used aggregate mortality as the health proxy for DI trends — a poor measure of mental health morbidity; accounting for the depression-physical illness interaction suggests physical illness's standalone causal role in disability has been overstated

Concepts Introduced or Extended

• Conditional DI Applicants — depression as a distinct trigger pathway: a mental health condition that raises both direct application rates and amplifies responses to physical illness and widowhood; not captured by aggregate-mortality health proxies
• DI Countercyclicality — tangential: Conti et al. identifies a health-shock channel (mental illness) that does operate for secular DI levels; this is compatible with Cutler et al. (2012/2015) rejecting health shocks for cyclical variation — depression prevalence is not countercyclical, so it explains trends without explaining recession surges

Entities Mentioned

• Rena M. Conti
• Ernst R. Berndt
• Richard G. Frank
• David Cutler (volume editor)
• David Wise (volume editor)
• Health and Retirement Study — uses CES-D (8-item, $\geq 4$ cutoff) and 39-condition health battery from waves 2–5; first systematic modeling of CES-D in HRS for DI application outcomes

Quotes

"Our results suggest that mental illness alone and in combination with physical illness exert important influences on retirement behavior and DI/SSI applications consistent with recent disability application and recipiency trends."

"Aggregate mortality statistics are likely correlated with the morbidity associated with acute and some chronic diseases but represent a rather imprecise measure of the health of the population."

My Take

Important for two reasons. First, it identifies depression as a quantitatively large DI application trigger that prior economics literature missed by using aggregate mortality as the health proxy. Second, the depression × widowhood interaction for men is counterintuitive — it suggests bereavement operates as a vulnerability amplifier, not a standalone trigger. The GEE estimation does not fully resolve endogeneity (depression and job loss are jointly determined), so the direct effect estimates should be read as upper bounds. The reconciliation with Cutler et al. (2012): both can be right simultaneously — depression explains secular DI level trends while not explaining cyclical variation, since depression prevalence does not spike in recessions.