The association of earnings with health in middle age: Do self-reported earnings for the previous year tell the whole story?

Research on earnings and health frequently relies on self-reported earnings (SRE) for a single year, despite repeated criticism of this measure. We use 31 years (1961-1991) of earnings recorded by the United States Social Security Administration (SSA) to predict the 1992 prevalence of disability, diabetes, stroke, heart disease, cancer, depression and death by 2002 in a subset of Health and Retirement Study participants (n = 5951). We compare odds ratios (ORs) for each health outcome associated with self-reported or administratively recorded earnings. Individuals with no 1991 SSA earnings had worse health in multiple domains than those with positive earnings. However, this association diminished as the time lag between earnings and health increased, so that the absence of earnings before approximately 1975 did not predict health in 1992. Among those with positive earnings, lengthening the lag between SSA earnings and health did not significantly diminish the magnitude of the association with diabetes, heart disease, stroke, or death. Longer lags did reduce but did not eliminate the association between earnings and both disability and depression. Despite theoretical limitations of single year SRE, there were no statistically significant differences between the ORs estimated with single-year SRE and those estimated with a 31-year average of SSA earnings. For example, a one unit increase in logged SRE for 1991 predicted a 19% reduction in the odds of dying by 2002 (OR = 0.81; 95% confidence interval: 0.72,0.90), while a similar increase in average SSA earnings for 1961-1991 had an OR of 0.72 (0.63, 0.82). The point estimates for the OR associated with 31 year average SSA earnings were further from the null than the ORs associated with single year SRE for heart disease, depression, and death, and closer to the null for disability, diabetes, and stroke, but none of these differences was statistically significant.