We estimate the effect of education on smoking. Our estimation strategy "differences out" the impact of unobserved characteristics correlated with education by exploiting education differences between similarly selected groups 1 year apart in their life cycle. Individuals with a given age,...