Autor, Kostøl, and Mogstad 2015 — Disability Benefits, Consumption Insurance, and Household Labor Supply

Summary

Uses Norwegian judge-lottery instrumental variables (IV) ($N = 14{,}077$, 75 judges, 1994–2005) to trace how disability insurance (DI) denial ripples through the full household — own earnings, benefit substitution from other programs, and the spousal Added Worker Effect (AWE) — then imputes consumption from administrative registers to recover DI's insurance value separately by marital status. For married couples, both household income and consumption are statistically unaffected by DI denial (full self-insurance via the AWE), yielding near-zero willingness-to-pay (WTP); for single applicants, denial produces large consumption drops, yielding WTP $\approx \$9{,}100$ per capita. The paper provides the first causal estimate of the DI-specific AWE using quasi-random assignment and the first household-level welfare analysis of DI.

Key Claims

1. Norwegian judge-lottery IV replicates the French-Song (2014) design in Norway: 75 administrative law judge (ALJ)-stage appeal judges randomly assigned within department $\times$ year cells; leave-one-out mean allowance rate as instrument; strong first stage. Setting: Norwegian National Insurance DI program, 1994–2005; $N = 14{,}077$ appellants.
2. Fiscal cost per DI allowance is $\approx \$16{,}000$ per year: gross benefit $\approx \$17{,}300$ per year minus $\approx 30$-cent benefit substitution per $\$1$ DI (Norwegian disability pension and sickness insurance partially offset), minus the labor supply channel.
3. Earnings effect — DI allowance reduces own earnings by $\approx \$6{,}600$ per year (year 1), stable across follow-up years; implies substantial residual work capacity among marginal Norwegian applicants.
4. Benefit substitution — roughly $30$ cents of other disability/sickness transfers are averted per $\$1$ DI received; the Norwegian safety net means DI's net fiscal cost is substantially below the gross benefit level.
5. Spousal Added Worker Effect — denied applicants' spouses earn $\approx \$6{,}000$ more in year 1 and $\$10{,}000$–$\$12{,}000$ more in years 2–4; first causal identification of the DI-AWE using quasi-random assignment (vs. Chen (2012)'s accepted/rejected comparison design).
6. Household income decomposition — four channels attenuate the gross DI shock: (1) own DI benefit; (2) benefit substitution from other programs; (3) own earnings change; (4) spousal AWE. Together these channels substantially reduce the net welfare loss from denial.
7. Marital status heterogeneity (income) — married: household income change $-\$1{,}615$ per capita (not significant); single: $+\$9{,}086$ per capita ($p < 0.01$). DI denial costs married households almost nothing in household income.
8. Marital status heterogeneity (consumption) — married: consumption change $-\$0.83$ per capita (not significant); single: $+\$9{,}835$ per capita ($p < 0.1$). Married households fully self-insure consumption against DI denial via the spousal AWE; single households cannot.
9. Benefit-to-cost ratios — married: $-0.37$ (DI receipt yields slightly negative net welfare relative to denial); single: $0.86$ (DI receipt yields large net welfare gains). These are per-dollar marginal welfare estimates.
10. Willingness to pay (static model) — cash-equivalent value of DI allowance at the margin: married $\approx \$2{,}700$ per capita; single $\approx \$9{,}100$ per capita. Single applicants value DI at $\approx 3.4\times$ the married rate.
11. Willingness to pay (dynamic model) — accounting for reapplication: married $\approx \$2{,}900$ per capita; single $\approx \$5{,}900$ per capita. Dynamic model compresses the gap somewhat.
12. Reapplication and AWE as offsetting mechanisms — together they offset $\approx 60\%$ of the welfare loss that DI denial would otherwise impose, confirming that DI functions primarily as consumption insurance for single households without internal risk-sharing.

Concepts Introduced or Extended

• Causal Effects of DI Receipt
• Added Worker Effect
• Nonhealth Risk and DI Insurance Value

Entities Mentioned

• David Autor
• Andreas Ravndal Kostøl
• Magne Mogstad
• Eric French
• Jae Song
• Susan Chen
• Hamish Low
• Luigi Pistaferri

Quotes

"We find that the insurance value of DI is substantially larger for single than for married applicants, reflecting that married applicants are better insured through the household."

"Taken together, reapplication and the spousal Added Worker Effect offset a large fraction of the welfare loss from DI denial."

My Take

The paper's main contribution is triangulating judge-lottery IV with consumption imputation from Norwegian administrative registers to reach a welfare analysis beyond earnings effects. The married/single heterogeneity is the most striking finding — married applicants have near-zero WTP for DI because the AWE provides full household-level self-insurance, while singles depend almost entirely on DI as their only consumption-smoothing mechanism. The fiscal cost accounting ($\approx \$16{,}000$ per year after benefit substitution) is more complete than prior US estimates that ignored program substitution. The AWE identification is cleaner than Chen (2012)'s comparison-group approach because it uses quasi-random judge assignment. Main caveat: Norwegian applicants (ALJ-stage, 1994–2005) differ from US in disability severity composition, household structure, and safety net generosity; whether the near-zero WTP for married couples generalizes is an open question.