b e rna rd b l ac k j o s é-a n ton i o e sp í n-s á n c h e z e ri c f re n c h kat e l i t va k A B S T R A CT We use the best available longitudinal data set, the Health and Retirement Study, and a battery of causal inference methods to provide both central estimates and bounds for the long-term effect of health insurance on health and mortality among the near-elderly (initial age 50-61) over a 20-year period. Compared with matched insured persons, those uninsured in 1992 consume fewer health-care services, but their health (while alive) does not deteriorate relative to the insured, and, in our central estimates, they do not die significantly faster than the insured. Our upper and lower bounds suggest that prior studies have greatly overestimated the health and mortality benefits of providing health insurance to the uninsured.