“…We control for both subjective and objective measures of baseline health, because both appear to affect individual's expectations about retirement (McGarry, 2004). Due to the possibility of justification bias from including subjective health, we create a predicted value of self-rated health in the first wave we observe an individual, using an ordered probit model (Erdogan-Ciftci, Van Doorslaer, and Lopez-Nicolas, 2008 Demographics. The models control for education (some college, college, missing education, with high school or under as the reference category), race (White, Black, Race Missing, with White being the reference category), and Hispanic ethnicity.…”