Presentation on theme: "Comments on Social Security Income and Elderly Mortality Tom Buchmueller UC-Irvine."— Presentation transcript:
Comments on Social Security Income and Elderly Mortality Tom Buchmueller UC-Irvine
Overview Fundamental question in health economics –Difficult inference problem Interesting study population –Simultaneity problem is reduced –Important from a policy perspective Clever identification strategy Nice data set –(but small samples)
Theory Income may influence health in a number of ways Which mechanism are likely to be most important for this population? 1 Possible Story: –Income increases prescr. drug use (Moran & Simon 2006) –Presc. drugs reduce mortality (Gowrisarkaran & Town 2004) Others?
The Experiment Women who were married at least 10 years and then divorced (and remain divorced) receive SS benefits based on exs earnings. Death of the ex increases SS benefits (if exs PIA > own PIA) Main comparison: divorced women whose ex dies with divorced women whose ex is alive. –Other possible: divorced after 9 yrs vs. divorced after 10 yrs. How do treatment and control groups compare to each other in terms of characteristics other than income?
Comparison with Snyder and Evans Similarity: both use SS rules that generate exogenous income variation. Differences: –Looks at women rather than men –Larger shock to income –Source of variation is not related to age What are the implications of these differences? What should we make of the differing results?
Data: New Beneficiary Data System Strengths –Accurate information on marital status –Matched with administrative files –Detailed controls Weakness –Small sample (768 divorced women)
Some Details Conceptual questions: Does it matter when they were divorced? Does it matter when he dies? Empirical question: What % of ex husband are dead? For divorcees, what % of retirement income is from SS? What does the first-stage look like?
Model and Results IV = 1[receives spousal benefits & ex is dead] –But receipt of spousal benefits is determined by own income vs. spouses income. OLS estimates are insignificant. –Shouldnt they have a negative bias? If so, what does this mean? IV results: zero or just imprecise?
Concluding Thoughts and Possible Extensions Probably a clean experiment, but convince me. Even if it is clean, power is a concern. Can analysis be done with administrative data? Are there other relevant outcomes in these data?