Summary
Using retrospective Survey of Income and Program Participation (SIPP) data merged to Social Security Administration (SSA) administrative earnings records, Singleton (2009) tests whether work-limiting disability onset raises the probability of divorce, and whether the effect is heterogeneous in theoretically predictable ways. Grounded in the Becker-Landes-Michael (1977) framework of marital instability, the paper shows that disability's effect on divorce is greatest when the disability is unanticipated (young workers, higher-educated) and when it most severely damages the couple's economic prospects (work-preventing severity). The headline result: among young, educated men whose disability completely prevents work, the divorce hazard is 13.3 percentage points (pp) higher than comparable nondisabled men, even after controlling for marital quality characteristics.
Key Claims
- Disability-divorce gradient by age: The effect of disability on divorce decreases with age. By ages 53–64, the effect is near zero and statistically insignificant. This is theoretically predicted because disability is more anticipated (higher baseline hazard) among older workers, so it generates less surprise to the marriage.
- Disability-divorce gradient by education: The effect increases with education. Young, high-education men see a +5.2 pp raw divorce hazard difference (any disability) and +20.2 pp for work-preventing disability only. The education gradient reflects both that disability is less anticipated among more educated workers (lower hazard) and that earnings losses are larger.
- Disability-divorce gradient by severity: Work-preventing disability (completely unable to work) raises divorce rates substantially more than work-limiting disability, consistent with the earnings-loss mechanism.
- After marital quality controls (Table 5): The divorce hazard difference for young, educated men with work-preventing disability is 13.3 pp — still large and statistically significant. The first-marriage indicator drives much of the attenuation.
- Female results: Similar qualitative patterns (decreasing with age, increasing with education) but smaller in magnitude and statistically insignificant throughout.
- Earnings losses (Table 2): Fixed-effects earnings regressions show young, educated men lose approximately $40,000 (2004 dollars) in earnings by year 7 post-onset. Among the 53–64 age group, the earnings effect is statistically insignificant — not because the disabled recover, but because nondisabled earnings also decline at that age.
- Reconciles Charles and Stephens (2004): That study found no aggregate disability-divorce effect. Singleton argues population-wide analysis obscures the effect because disability is concentrated among older workers where the divorce effect is smallest.
- Disability Insurance (DI) policy implication: More generous formal disability insurance could reduce divorce (by replacing lost earnings) or increase it (by raising the outside-marriage option). The net effect is an open empirical question.
Concepts Introduced or Extended
Entities Mentioned
Quotes
"This study shows that disability not only affects divorce in some cases, but does so in a manner consistent with theoretical predictions."
"Formal disability insurance may actually increase divorce rates by making the alternatives to marriage more attractive: disability benefits may discourage work and encourage divorce among individuals who would otherwise work and remain married."
My Take
The paper's core strength is its tight theoretical discipline: it derives heterogeneous predictions from a formal model and confirms all of them in SIPP data. The reconciliation of the Charles-Stephens null result is particularly elegant — the aggregate null is consistent with strong subgroup effects when the high-effect subgroup (young educated men) is a small share of the disabled population. The identification is observational (no instrumental variable (IV) for disability onset), so estimates capture association rather than clean causation; marital quality controls are partial (first-marriage indicator only). The female null result is consistent with the theory's predictions given smaller earnings gradients, but the contribution of social stigma (divorcing a disabled woman) versus economics remains unidentified.