Abstract: In the first chapter, I study whether sending children to college affects the contemporaneous labor supply of their parents. I use data on the retirement behavior of parents and their children's college enrollments from the Health and Retirement Survey and show that parents do delay retirement when they are paying for their children to attend college. Mothers and fathers are more likely to be working, less likely to be collecting Social Security benefits, and less likely to report that they are retired if they are currently paying for the college education of a child. For those who continue working, I find little evidence of any impact on work intensity. For fathers, the pattern of effects is consistent with a model of precautionary savings. In the second and third chapters, I study whether the Social Security Notch can be used as a source of exogenous variation in post-retirement incomes. Using a synthetic cohorts approach, I model Social Security Benefits for people born during 1905-1935, showing what retirement benefits for each cohort would be if changes were due only to differences in benefit calculation formulae. I show that these simulated benefits are very significant predictors of reported benefits and of reported total incomes in very large data sources, such as the Current Population Survey and the Census. However, the cohorts with higher benefits due to law changes are also observed to have higher earnings in retirement. I then examine whether the notch can be used to discern a causal relationship between post-retirement income and health. I replicate the work of Snyder and Evans, showing that people born in 1917 received lower Social Security benefits in retirement than people born in 1916, yet had lower mortality rates after age 65. Using data from the Death Master File on pre-65 mortality rates, however, I find that gap in mortality rates appeared before these cohorts began collecting Social Security benefits. I also find that the difference in mortality rates between these cohorts is similar in magnitude to the gaps between other successive cohorts. These results imply that cohort variability overshadows the effects of this potential instrument.