Burkhauser and Cawley 2004 — Obesity, Disability, and Movement Onto the Disability Insurance Rolls

Summary

Burkhauser and Cawley (2004) use two-stage least squares (2SLS) with biological relatives' weight as an instrument for own body mass index (BMI) to estimate the causal effect of obesity on self-reported work limitations and disability income receipt among working-age adults (ages 25–44) in the Panel Study of Income Dynamics (PSID) and National Longitudinal Survey of Youth 1979 (NLSY79). Ordinary least squares (OLS) and linear probability estimates are severely downward-biased relative to 2SLS — by a factor of $5$–$15\times$ — due to measurement error in BMI and endogeneity from reverse causation (disability reducing mobility and raising weight). The 2SLS estimates show obesity causally raises disability income receipt by roughly $7$–$9$ percentage points (pp) for men (NLSY) and women (PSID and NLSY). The paper contextualizes these findings in the parallel doubling of both obesity rates and Disability Insurance (DI) rolls since the early 1980s, and documents the Social Security Administration's (SSA) administrative history of the obesity Listing — deleted in 1999, restored via Social Security Ruling (SSR) 02-1p in September 2002.

Key Claims

• OLS and linear probability models underestimate the causal effect of obesity on disability outcomes by $5$–$15\times$, primarily due to attenuation bias from BMI measurement error and endogeneity (disability reducing physical activity and raising weight)
• 2SLS instruments — weight of a biological sibling (NLSY) or adult child/parent (PSID) — have strong first stages: $F$-statistics range from $38$ to $99$
• Men (NLSY, 2SLS): obesity → $+5.38$ pp work limitations ($10\%$ sig.); $+6.92$ pp disability income receipt ($1\%$ sig.)
• Women (PSID, 2SLS): obesity → $+9.21$ pp disability income receipt ($5\%$ sig.)
• Women (NLSY, 2SLS): obesity → $+5.64$ pp disability income receipt ($1\%$ sig.)
• Two-year disability income receipt: only NLSY women significant — obesity → $+3.67$ pp; other groups insignificant (insufficient power or PSID sample limitations)
• Obesity rates roughly doubled between the early 1980s and 2002; Supplemental Security Income (SSI) / DI rolls also roughly doubled over the same period — the trends are correlated
• SSA deleted obesity as a stand-alone Listing of Impairments entry in 1999 (the existing criteria were deemed inadequate) and restored it via SSR 02-1p (Federal Register, September 2002) with new evaluation criteria
• The exclusion restriction holds because biological relatives' weight operates through shared genetic predisposition, not through shared behavioral/environmental pathways that would independently affect the respondent's disability

Concepts Introduced or Extended

• DI Growth Decomposition — obesity as a health-demand driver of the adjusted incidence residual; the post-1980 obesity doubling contributes to rising musculoskeletal conditions
• Listing of Impairments — SSA obesity listing history: deleted 1999, restored via SSR 02-1p September 2002
• Morbidity-Mortality Distinction — obesity is a working-age morbidity driver that raises DI receipt without necessarily raising short-run mortality
• Conditional DI Applicants — obesity creates a pool of health-impaired workers who may become conditional DI applicants upon job loss

Entities Mentioned

• Richard V. Burkhauser
• John Cawley

Quotes

"OLS and logit estimates of the impact of obesity on these outcomes are severely downward biased."

"A one-unit increase in the probability of being obese leads to a [5–9 percentage point] increase in the probability of disability income receipt."

My Take

Strong IV design and the OLS-to-2SLS divergence is striking and methodologically important — the paper establishes that conventional estimates massively understate the obesity-disability link. However, results are mixed: several subgroup estimates are insignificant, the NLSY disability income variable is household-level and may capture a spouse's income, and the sample is limited to ages 25–44 (DI is most prevalent at 50–64). The paper remains a working paper (MRRC 2004-089), not peer-reviewed. Despite these limitations, it is the first IV-based study of obesity-to-DI causality and provides the clearest available evidence that the doubling of obesity rates since the 1980s contributed causally — not merely correlationally — to rising DI receipt.