Chen and van der Klaauw 2008 — The Work Disincentive Effects of the Disability Insurance Program in the 1990s

Summary

Chen and van der Klaauw update Bound's (1989) comparison-group upper bound on Disability Insurance (DI) work disincentives using 1990s Survey of Income and Program Participation (SIPP)–Social Security Administration (SSA) merged data, finding at most $20$ percentage points (pp) labor force participation (LFP) reduction — substantially smaller than Bound's $30$ pp. They also introduce a fuzzy regression-discontinuity (RD) design exploiting the vocational grid's age cutoffs at 45, 50, and 55, which generates cleaner causal identification for the subset of Stage 5 (vocational) applicants. Effects for this marginal vocational group are even smaller ($\approx 11$–$16$ pp), consistent with the interpretation that applicants whose eligibility depends on age-based vocational factors retain more residual work capacity. The paper extends Bound's male–Social Security Disability Insurance (SSDI) framework to women and Supplemental Security Income (SSI) applicants.

Key Claims

• Comparison-group upper bound: At most $14$–$15$ pp LFP reduction (unconditional); $19$–$21$ pp after correcting for $\approx 24\%$ misclassification of rejected applicants who were actually allowed on subsequent appeals. Both are smaller than Bound's (1989) $30$ pp.
• Stage 5 vocational subgroup: Unconditional LFP effect $-14.62$ (short-run) to $-15.65$ (long-run); conditional on demographics $-11.07$ to $-11.16$ pp. Smaller than the all-applicant estimates.
• Fuzzy RD: The vocational grid creates discontinuities in the award probability at ages 45, 50, and 55. Award rates are not deterministic at cutoffs (unobserved residual functional capacity (RFC), illiteracy, and administrator discretion are also factors), making the design fuzzy rather than sharp. The size of each discontinuity depends on the fraction of applicants around each cutoff for whom age can influence the grid determination.
• Long-run ≈ short-run: The work-disincentive estimates are stable from 1–24 months to 1–11 years post-decision, providing evidence against the hypothesis that the disincentive is primarily a transient earnings-suppression effect during appeals.
• Misclassification bias: $\approx 24\%$ of "rejected" applicants in the 831 file are actually misclassified beneficiaries (allowed at Administrative Law Judge (ALJ) or higher, which the 831 file does not capture). Correcting for this multiplies estimates by $1.32$; even the bias-adjusted estimates ($19$–$21$ pp) remain smaller than Bound.
• SSI vs. SSDI: SSI applicants have lower pre-application LFP and earnings than SSDI applicants; including them mechanically reduces the comparison-group bound. Rejected SSI applicants are less likely to work than rejected SSDI applicants (Bound et al. 2003), consistent with SSI applicants having weaker labor force attachment independent of benefit receipt.
• Gender: The analysis extends to women; female applicants show smaller disincentive effects than male applicants, consistent with their lower pre-application LFP. SSDI male applicants show the largest effects within the Stage 5 vocational subgroup.

Concepts Introduced or Extended

• Causal Effects of DI Receipt
• Vocational Grid
• SSA Sequential Determination Process
• DI Denied Population
• Conditional DI Applicants
• Substantial Gainful Activity

Entities Mentioned

• Susan Chen
• Wilbert van der Klaauw
• Social Security Administration
• Disability Insurance Program

Quotes

"Our estimates indicate that the work disincentive effects associated with DI benefit receipt during the 1990s were relatively modest, implying that the LFP rate of DI beneficiaries would have been at most 20 percentage points higher had they not received benefits."

"We find even smaller effect estimates for a group of 'marginal' applicants whose disability determination was based on vocational considerations."

"Instead of being a deterministic function of age, the award decision made at stage 5 of the application process can be characterized by a probability function that will be discontinuous in age at the three cutoff values found in the grid at ages 45, 50 and 55."

My Take

The paper's main contribution is methodological: it demonstrates that the vocational grid's age cutoffs can serve as a source of quasi-random variation for a policy-relevant subset of applicants. This is a direct precursor to Maestas, Mullen, and Strand (2013)'s examiner-IV design, which generalizes quasi-random assignment to all applicants rather than just the Stage 5 vocational subset. The comparison-group results (smaller than Bound) are best interpreted as an update for the 1990s rather than a causal estimate — the SSI inclusion and gender extension mechanically reduce the bound, and the paper is transparent about this. The stability of short-run vs. long-run estimates is a useful diagnostic, though the 831 file's inability to track appeals beyond reconsideration means the "rejected" comparison group is noisily defined.