The association between the mortality rate and individual characteristics such as social class, income, education, and gender is different at different ages. This is often interpreted as evidence that the causal effects at the individual level change with age. However, in the presence of unobserved heterogeneity, dynamic selection may occur at a higher speed among subjects with observed characteristics that increase the mortality rate, and this will also, by itself, generate age variation in the association between characteristics and mortality. We show that these explanations can be distinguished empirically. We derive a simple test and we apply it to data on mortality among the elderly from the US National Longitudinal Mortality Study. We find that dynamic selection explains observed mortality variation across age for many covariates, but selection can not explain the large relative mortality improvement among black men at high ages. We extend our approach to the macro level, using country as the unit of observation. Dynamic selection explains the attenuation of the association of GDP and mortality at high ages.

Please sign up for meetings using the below:
docs.google.com/spreadsheets/d/1HspRWfvwCp5ocoqmoZUZfigDfdiwf4hzjxoCdHEse4c/edit#gid=0