Working-paper note: This is the 2011 dissertation/working-paper version. The published version is Lindner 2016 — How Do Unemployment Insurance Benefits Affect the Decision to Apply for Social Security Disability Insurance (Journal of Human Resources 51(1): 62–94), which extends the data window to 2007, switches to Cox proportional hazard plus spell-level logit models (diagnosing unobserved heterogeneity), and adds an optimal-UI-formula extension. This page is retained for its distinct semi-parametric correlated-random-effects methodology and its insurance-vs-search-effort framing.
Lindner (2011) asks whether unemployment insurance (UI) benefits affect the decision to apply for disability insurance (DI), and which of two theoretically competing channels dominates: the insurance channel (UI substitutes for DI cash income, reducing the marginal value of applying) or the search effort channel (DI applicants search less for work, so higher UI delays but ultimately raises DI applications among low-searchers). Using Survey of Income and Program Participation (SIPP) data 1990–2004 matched to Social Security Administration (SSA) administrative records and a discrete-time proportional hazard model with semi-parametric correlated random effects (Heckman-Singer 1984), he finds that higher UI monthly benefits significantly reduce the hazard to DI application among UI recipients — the insurance channel dominates. UI take-up itself also significantly reduces DI application once correlated costs of both program entry are accounted for via random effects.
"My empirical results indicate that higher UI benefits do indeed reduce the hazard to DI application. The quantitative effect is also sizable for those who receive UI benefits."
"Application decisions for DI appear to be much more sensitive to even short-term cash incentives than has been previously acknowledged, at least among health impaired workers who have lost their jobs and who rely on short-term cash assistance such as UI benefits."
The paper's central identification strategy — using cross-state variation in UI benefit formulas to identify the causal effect on DI applications — is credible and the theoretical framework is unusually complete for a reduced-form paper of this era. The Heckman-Singer correction for correlated UI/DI take-up costs is a genuine methodological contribution: without it, the positive correlation between UI take-up and DI application (both require low stigma) biases the coefficient toward zero or positive, and the insurance channel effect is obscured.
The main limitation is weak first-stage instruments for UI take-up (F-statistic ≈ 3.5), which compromises the local average treatment effect (LATE) interpretation of the UI take-up effect. The benefit level effect (identified from cross-state variation in the benefit formula, controlling for prior earnings) is better identified and more interpretable.
The null result from Mueller, Rothstein, and von Wachter (2016) — who find no UI-SSDI interaction using a later sample and a different identification approach — is not addressed, since it postdates this paper. The discrepancy is unresolved: possible explanations include different time periods (1990–2004 vs. 2000s–2010s), different identification strategies, or sample composition differences (Lindner restricts to UI-eligible, DI-eligible, work-limited job losers, a much narrower population). This paper's narrow sample is its strength for cleanly identifying the theoretical channels, but it limits external validity for general UI-SSDI policy interactions.
The cost-effectiveness calculation is useful back-of-envelope but relies on strong assumptions: that the sample's application behavior generalizes to all DI applicants, and that the PDV calculation from von Wachter et al. (2010) is applicable.