Autor and Duggan 2007 — Distinguishing Income from Substitution Effects in Disability Insurance

Summary

Six-page AEA Papers and Proceedings paper introducing the income vs. substitution effect distinction for disability insurance (DI) labor supply, and presenting preliminary evidence using the Veterans' Disability Compensation (VDC) program as a natural experiment. VDC benefits are not work-contingent, so any labor force participation (LFP) reduction caused by receiving them is attributable to the pure income effect. The 2001 Agent Orange ruling added diabetes to VDC qualifying conditions for Vietnam veterans, generating a quasi-random income shock. Difference-in-differences (DiD) estimates on Current Population Survey (CPS) data show a $3.21$ percentage points (pp) not in labor force (NILF) increase for affected veterans relative to nonveteran controls, split roughly equally between retirement and disability channels — despite the absence of any implicit tax on work.

Key Claims

• Two channels, one policy target: Social Security Disability Insurance (SSDI) discourages work through (1) a substitution effect — benefits are lost if earnings exceed substantial gainful activity (SGA) ("cash cliff"), creating an implicit tax on labor — and (2) an income effect — permanent income and Medicare may make early retirement preferred even if work is not taxed. All prior Congressional and SSA interventions (Ticket to Work, trial work periods, Medicare continuation eligibility) have targeted only the substitution effect.
• VDC as pure income instrument: VDC benefits are paid regardless of employment status and are not means-tested. Any reduction in labor supply caused by VDC receipt cannot operate through the substitution channel — it must be a pure income effect.
• Agent Orange natural experiment: In 2001, the Veterans Affairs (VA) added diabetes to the list of qualifying conditions for Vietnam veterans with "boots on ground" exposure. This generated $\approx 175{,}000$ new VDC enrollees ($+2.3\%$ of living Vietnam-era veterans) and $\approx 75{,}000$ benefit increases, accelerating annual VDC enrollment growth from $0.6\%$ to $3.2\%$ per year.
• DiD main result: Vietnam-era veteran (VEV) $\times$ Post $= +3.21$ pp NILF ($p < 0.05$) for VEV relative to nonveteran males born 1941–1952, after 2001. This is a large LFP reduction with no implicit work disincentive operating.
• Decomposition: $+1.77$ pp NILF-Retired, $+1.31$ pp NILF-Disabled; $+0.12$ pp NILF-Other (not significant). Early retirement and disability channels absorb the LFP reduction in roughly equal shares.
• Self-reported disability: $+1.30$ pp ($p < 0.05$). Authors flag this may partly reflect reporting behavior rather than health change (cf. Bound and Waidmann 1992).
• Ticket to Work (TTW) failure rate: Fewer than $1{,}400$ of $12.2$ million TTW tickets issued led to successful workforce integration ($0.01\%$). Consistent with the income-effect interpretation: the program eliminates the substitution disincentive but cannot address beneficiaries' preference for early retirement financed by transfer income.
• Caveats: Nonveteran control group differs observably from Vietnam-era veterans ($4\%$ vs. $15\%$ high-school dropout rate; $28\%$ vs. $35\%$ college graduates). Results are explicitly preliminary; the companion paper (Autor and Duggan 2007, unpublished) exploits military service records for sharper identification.

Concepts Introduced or Extended

• Causal Effects of DI Receipt — adds VDC natural experiment as fifth identification strategy for income vs. substitution decomposition; reframes Ticket to Work critique through income-effect lens

Entities Mentioned

• David Autor
• Mark Duggan
• Social Security Administration
• Disability Insurance Program

Quotes

"Our core observation is that SSDI, and indeed all nonwork-contingent retirement programs, discourage work through two channels. The first is the canonical substitution effect... The second is the income effect—given the transfer payments and in-kind services (particularly medical care) provided by SSDI, many beneficiaries may prefer leisure to labor or, more precisely, an early retirement, even if work is not implicitly taxed by the SSDI program."

"If, however, the primary means by which SSDI reduces labor force participation and hastens retirement is through an income effect, such efforts may be close to ineffectual."

"While economists have typically regarded the substantial reductions in labor force participation associated with receipt of disability benefits as an incentive problem (i.e., a substitution effect), it appears plausible to us that a significant share of this response is explained by the (nonincentive) income effect."

My Take

This paper is best read as the conceptual framing for a research agenda rather than a definitive empirical result. The identification is creative — VDC as a non-work-contingent analog to SSDI — and the Agent Orange ruling provides genuine exogenous variation. The DiD limitation is real (nonveterans differ observably), and the authors are honest about it. The lasting contribution is the reframing: all prior work-incentive policy addressed only the substitution channel. The income-effect lens explains both the TTW failure ($0.01\%$ success) and the pattern documented by Gelber, Moore, and Strand (GMS) 2016: their regression kink design (RKD) finds that $\$1$ of DI income reduces earnings by $\$0.20$ with no SGA-threshold mechanism — pure income effect. AD (2007) conceptually predicted what GMS (2016) subsequently identified rigorously. The two papers, seven years apart, triangulate to the same conclusion: DI's work disincentive is primarily income-driven, and substitution-targeted policy is attacking the wrong channel. This has a direct implication for continuing disability review (CDR) design: removing benefits eliminates both channels, while benefit reduction (via RKD variation) operates only on income effects. The optimal policy design question — how to target the substitution channel without triggering income-effect–driven LFP reduction — remains open.