Bound Lindner and Waidmann 2014 — Reconciling Findings on the Employment Effect of Disability Insurance

Summary

Bound, Lindner, and Waidmann (2014) reconcile a long-standing tension between two approaches to measuring Disability Insurance (DI)'s employment effect: aggregate time-series studies attributing most of the post-1980 employment decline among work-limited men to DI expansion, and instrumental variables (IV) / comparison-group studies (Bound 1989, French and Song 2014, von Wachter et al. 2011) finding more modest point estimates of $15$–$30$ percentage points (pp). Using the Survey of Income and Program Participation (SIPP) matched to Social Security Administration (SSA) administrative records, the authors decompose the employment changes of work-limited men across three population bins — DI beneficiaries, denied applicants, and non-applicants with work limitations — at three snapshots (1990, 1996, and 2004). They find that during the rapid-expansion early 1990s, DI could explain $50$–$100\%$ of the employment decline under extreme assumptions, but during the mid-1990s to mid-2000s, when DI rolls barely grew, employment among work-limited men continued to fall, implying DI accounts for at most $20$–$50\%$. The conclusion is that aggregate studies overstate DI's causal role by conflating program expansion with concurrent secular declines, while the IV estimates are internally consistent with each other.

Key Claims

• During the rapid DI expansion of 1990–96, DI can account for $50$–$100\%$ of the employment decline among work-limited men — but only under extreme assumptions about their counterfactual work capacity; the range reflects sensitivity to what fraction of beneficiaries would have worked absent DI.
• During 1996–2004, DI enrollment growth was minimal while employment among work-limited men continued to decline; DI explains at most $20$–$50\%$ of this later decline. Other factors — secular health, labor demand, compositional shifts — are necessary.
• The tension between aggregate and IV estimates is not a contradiction: aggregate approaches attribute all co-occurring employment decline to DI (an overstatement), while IV estimates identify a well-defined partial-equilibrium causal effect. The true causal estimate is closer to the IV range.
• The three-bin decomposition (beneficiaries / denied / non-applicants) in SIPP-SSA data shows that the ratio of beneficiaries to work-limited non-applicants rose substantially in the early 1990s but leveled off thereafter, consistent with the aggregate pattern.
• Bound's (1989) canonical comparison-group upper bound ($\approx 30$ pp) and French and Song's (2014) Administrative Law Judge (ALJ) judge-IV estimate ($\approx 26$–$28$ pp) are consistent once the aggregate overstatement problem is diagnosed.

Concepts Introduced or Extended

• Causal Effects of DI Receipt
• DI Denied Population

Entities Mentioned

• John Bound
• Stephan Lindner
• Timothy Waidmann
• Eric French
• Jae Song
• Till von Wachter
• Joyce Manchester

Quotes

"The trend in employment of men with work-impairments seems to be mostly driven by changes in the behavior of individuals who never apply for DI benefits rather than by DI itself."

"During the 1990–1996 period... the DI program could potentially explain the entire decline in employment for all work-impaired men. However, this would require extreme assumptions."

My Take

The paper's main contribution is diagnostic rather than estimating a new causal effect: it shows that the apparent conflict between aggregate-level and microeconometric estimates of DI's employment effect is a methodological artifact. Aggregate attribution is not the same as causal estimation. The decomposition framework is a useful check on how much macro-level employment trends can legitimately be attributed to a particular program. The finding that most of the employment decline among work-limited men in the late 1990s and early 2000s is not accounted for by DI growth is an underappreciated corrective to policy discussions that treat DI as the primary driver of declining male labor force participation.